Improved artificial intelligence or machine learning localization

Abstract

Example embodiments of the present disclosure relate to a solution for improving artificial intelligence (AI) or machine learning (ML) positioning. In one aspect, a terminal device sends a request to a network device for neural network modules cascaded after a pre-trained model. A terminal device receives, from a network device, an order of trained neural network modules cascaded after a pre-trained model and a set of input signals from a set of transmit receive points (TRP) to be input to the terminal device. The terminal device performs AI / ML positioning by using the set of input signals from the set of TRP in the above order based on trained neural network modules cascaded after the pre-trained model. In this way, tradeoff between model performance, model complexity, and computational complexity may be achieved.

Description

Technical Field

[0001] Various example embodiments relate to the field of communications, and in particular to devices, methods, apparatuses, and computer-readable storage media for improving artificial intelligence (AI) or machine learning (ML) positioning. Background Technology

[0002] From a standardization perspective, 3GPP Release 18 has approved research into the benefits of enhancing air interfaces through the latest advances in artificial intelligence (AI) or machine learning (ML), and AI / ML positioning accuracy enhancement is considered one of the three most representative use cases, such as positioning accuracy enhancement for different scenarios, including those with severe NLOS conditions. AI / ML positioning (e.g., direct AI / ML positioning) is considered an important area for further research. Therefore, improvements to AI / ML positioning are needed. Summary of the Invention

[0003] Overall, the exemplary embodiments of this disclosure provide a solution for improving AI / ML localization, such as direct AI / ML localization.

[0004] In a first aspect, a terminal device is provided. The terminal device may include: at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the terminal device to at least: send a request to a network device for a neural network module cascaded after a pre-trained model; receive from the network device a sequence of the trained neural network module cascaded after the pre-trained model and a set of input signals from a set of transmit-receive points (TRPs) to be input to the terminal device; and perform artificial intelligence (AI) / machine learning (ML) localization based on the trained neural network module cascaded after the pre-trained model, by using the set of input signals from the TRP set in the aforementioned sequence.

[0005] In a second aspect, a network device is provided. The network device may include: at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the network device to at least: receive from an end device a request for a neural network module cascaded after a pre-trained model; train the neural network module based on data from a set of Transmitter Receive Points (TRPs) selected by the end device; and send to the end device the trained neural network module cascaded after the pre-trained model, and a set of input signals from the TRP set, wherein the set of input signals from the TRP set is to be input into the pre-trained model cascaded with the trained neural network module at the end device.

[0006] In a third aspect, a method is provided. This method may include: sending a request to a network device for a neural network module cascaded after a pre-trained model; receiving from the network device the sequence of the trained neural network module cascaded after the pre-trained model and an input signal set from a set of transmit-receive points (TRPs) to be input to a terminal device; and performing artificial intelligence (AI) / machine learning (ML) localization based on the trained neural network module cascaded after the pre-trained model, by using the input signal set from the TRP set in the aforementioned sequence.

[0007] In a fourth aspect, a method is provided. This method may include: receiving from a terminal device a request for a neural network module cascaded after a pre-trained model; training the neural network module based on data from a set of transmit-receive points (TRPs) selected by the terminal device; and sending to the terminal device the trained neural network module cascaded after the pre-trained model, and a set of input signals from the TRP set, wherein the set of input signals from the TRP set is to be input into the pre-trained model cascaded with the trained neural network module at the terminal device.

[0008] In a fifth aspect, an apparatus is provided. The apparatus may include: components for sending a request to a network device for a neural network module cascaded after a pre-trained model; components for receiving from the network device the trained neural network module cascaded after the pre-trained model and a set of input signals from a set of transmit receiving points (TRPs) to be input to a terminal device in the order described above; and components for performing artificial intelligence (AI) / machine learning (ML) localization based on the trained neural network module cascaded after the pre-trained model by using the set of input signals from the TRP set in the aforementioned order.

[0009] In a sixth aspect, an apparatus is provided. The apparatus may include: means for receiving from a terminal device a request for a neural network module cascaded after a pre-trained model; means for training the neural network module based on data from a set of transmit receiving points (TRPs) selected by the terminal device; and means for sending to the terminal device the trained neural network module cascaded after the pre-trained model, and the sequence of an input signal set from the TRP set, wherein the input signal set from the TRP set is to be input into the pre-trained model cascaded with the trained neural network module at the terminal device.

[0010] In a seventh aspect, a non-transitory computer-readable medium is provided, comprising program instructions that, when executed by a device, cause the device to perform at least the method according to the third aspect.

[0011] In an eighth aspect, a non-transitory computer-readable medium is provided that includes program instructions, which, when executed by a device, cause the device to perform at least the method according to the fourth aspect.

[0012] In a ninth aspect, a computer program including instructions, when executed by a device, causes the device to at least: send a request to a network device for a neural network module cascaded after a pre-trained model; receive from the network device a sequence of a trained neural network module cascaded after a pre-trained model and a set of input signals from a set of transmit receiving points (TRPs) to be input to a terminal device; and perform artificial intelligence (AI) / machine learning (ML) localization based on the trained neural network module cascaded after the pre-trained model, by using the set of input signals from the set of TRPs in the aforementioned sequence.

[0013] In a tenth aspect, a computer program including instructions, when executed by an apparatus, causes the apparatus to at least: receive from a terminal device a request for a neural network module cascaded after a pre-trained model; train the neural network module based on data from a set of transmit receiving points (TRPs) selected by the terminal device; and send to the terminal device the trained neural network module cascaded after the pre-trained model and a set of input signals from the TRP set, wherein the set of input signals from the TRP set is to be input into the pre-trained model cascaded with the trained neural network module at the terminal device.

[0014] In an eleventh aspect, a terminal device is provided. The terminal device may include: a transmitting circuitry configured to send a request to a network device for a neural network module cascaded after a pre-trained model; a receiving circuitry configured to receive from the network device the trained neural network module cascaded after the pre-trained model and a set of input signals from a set of transmit receiving points (TRPs) to be input to the terminal device in that order; and an execution circuitry configured to perform artificial intelligence (AI) / machine learning (ML) localization based on the trained neural network module cascaded after the pre-trained model, by using the set of input signals from the TRP set in the order described above.

[0015] In a twelfth aspect, a network device is provided. The network device may include: a receiving circuitry configured to receive from a terminal device a request for a neural network module cascaded after a pre-trained model; a training circuitry configured to train the neural network module based on data from a set of transmit receiving points (TRPs) selected by the terminal device; and a transmitting circuitry configured to transmit to the terminal device, in sequence, the trained neural network module cascaded after the pre-trained model and a set of input signals from the TRP set, wherein the set of input signals from the TRP set is to be input into the pre-trained model cascaded with the trained neural network module at the terminal device.

[0016] It should be understood that the summary portion is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0017] Some exemplary embodiments will now be described with reference to the accompanying drawings, in which:

[0018] Figure 1 The illustration shows an example of a network environment in which some embodiments of the present disclosure may be implemented;

[0019] Figure 2 The diagram illustrates a signaling process for performing direct AI / ML localization according to some embodiments of the present disclosure;

[0020] Figure 3 The illustration shows an example signaling process according to some embodiments of the present disclosure for deploying a custom neural network (NN) module cascaded after a frozen TRP-agnostic general basis (TAB) model on the terminal device side for direct AI / ML localization;

[0021] Figure 4 The illustration shows example implementations of training a general model according to some embodiments of the present disclosure;

[0022] Figure 5 The illustration shows an example implementation of a lightweight custom neural network (NN) module cascaded after a frozen TAB model, according to some embodiments of the present disclosure;

[0023] Figure 6 The illustration shows a flowchart of an example method implemented at a terminal device according to some embodiments of the present disclosure;

[0024] Figure 7 The illustration shows a flowchart of an example method implemented at a network device according to some embodiments of the present disclosure;

[0025] Figure 8The illustration shows a simplified block diagram of an apparatus suitable for implementing some embodiments of the present disclosure; and

[0026] Figure 9 A block diagram illustrating an example of a computer-readable medium according to some embodiments of the present disclosure is shown.

[0027] Throughout the accompanying drawings, the same or similar reference numerals denote the same or similar elements. Detailed Implementation

[0028] The principles of this disclosure will now be described with reference to some exemplary embodiments. It should be understood that these embodiments are described for illustrative purposes only and to assist those skilled in the art in understanding and implementing this disclosure, and do not constitute any limitation on the scope of this disclosure. The disclosure described herein can be implemented in various other ways besides those described below.

[0029] In the following description and claims, unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.

[0030] In this disclosure, references to "an embodiment," "embodiment," and "example embodiment," etc., indicate that the described embodiment may include a particular feature, structure, or characteristic, but not every embodiment must include that particular feature, structure, or characteristic. Furthermore, such phrases do not necessarily refer to the same embodiment. Moreover, when a particular feature, structure, or characteristic is described in connection with an embodiment, those skilled in the art will understand that, whether explicitly described or not, combining it with other embodiments to affect such a feature, structure, or characteristic is within the knowledge of those skilled in the art.

[0031] It should 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 used only to distinguish one element from another. For example, without departing from the scope of the exemplary embodiments, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element. 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 limit the exemplary embodiments. The singular forms “a,” “an,” and “the” used herein also include the plural forms unless the context clearly indicates otherwise. Further understanding, the terms “includes,” “including,” “has,” “having,” “including,” and / or “including” as used herein specify the presence of the stated features, elements, and / or components, but do not exclude the presence or addition of one or more other features, elements, components, and / or combinations thereof. As used herein, “at least one of the following: ” and “<at least one item in a list of two or more elements>” and similar wording (where a list of two or more elements is connected by “and” or “or”) means at least any one of these elements, or at least any two or more of these elements, or at least all of these elements.

[0033] As used in this application, the term "circuit system" may refer to one or more or all of the following: (a) Pure hardware circuit implementation (such as implementation using only analog and / or digital circuit systems), and (b) A combination of hardware circuitry and software, such as (if applicable): (i) A combination of (multiple) analog and / or digital hardware circuits and software / firmware, and (ii) Any part of a hardware processor (including multiple digital signal processors), software, and memory (multiple processors) having software, which work together to enable a device (such as a mobile phone or server) to perform various functions, and (c) (Multiple) hardware circuits and / or (multiple) processors, such as (multiple) microprocessors or a portion thereof, which require software (e.g., firmware) to operate, but may not exist when operation is not required.

[0034] The definition of "circuit system" applies to all uses of the term in this application, including in any claim. As another example, as used in this application, the term "circuit system" also covers implementations of only hardware circuitry or processors (or processors in general) or portions thereof and their accompanying software and / or firmware. For instance, if applicable to a particular claim element, the term "circuit system" also covers baseband integrated circuits or processor integrated circuits for mobile devices, or similar integrated circuits in servers, cellular network devices, or other computing or networking devices.

[0035] As used herein, the term "communication network" refers to a network that conforms to any suitable communication standard, such as Long Term Evolution (LTE), LTE-A Advanced (LTE-A), Wideband Code Division Multiplexing (WCDMA), High-Speed ​​Packet Access (HSPA), Narrowband Internet of Things (NB-IoT), etc. Furthermore, communication between terminal devices and network devices in a communication network can be performed according to any suitable generation of communication protocol, including but not limited to first-generation (1G), second-generation (2G), 2.5G, 2.75G, third-generation (3G), fourth-generation (4G), 4.5G, future fifth-generation (5G) communication protocols, and / or any other protocols currently known or to be developed in the future. Embodiments of this disclosure can be applied to various communication systems. Given the rapid development of communications, there will naturally be communication technologies and systems of future types that can embody this disclosure. This should not be construed as limiting the scope of this disclosure to the systems described above.

[0036] As used herein, the term "network device" refers to a node in a communication network through which terminal devices access the network and receive services. Network devices can refer to base stations (BS) or access points (APs), such as Node B (NodeB or NB), evolved Node B (eNodeB or eNB), NR NB (also known as gNB), Remote Radio Unit (RRU), Radio Header (RH), Remote Radio Header (RRH), relay, low-power nodes (such as femtoseconds, picoseconds), etc., depending on the terminology and technology used.

[0037] The term "terminal device" refers to any terminal device capable of wireless communication. As an example and not a limitation, a terminal device may also be referred to as a communication device, user equipment (UE), subscriber station (SS), portable subscriber station, mobile station (MS), or access terminal (AT). Terminal devices can include, but are not limited to, mobile phones, cellular phones, smartphones, Voice over IP (VoIP) phones, wireless local loop phones, tablets, wearable terminal devices, personal digital assistants (PDAs), portable computers, desktop computers, image capture terminal devices (such as digital cameras), gaming terminal devices, music storage and playback devices, in-vehicle wireless terminal devices, wireless endpoints, mobile stations, laptop embedded devices (LEE), laptop mounted devices (LME), USB dongles, smart devices, wireless customer premises equipment (CPE), Internet of Things (IoT) devices, watches or other wearable devices, head-mounted displays (HMDs), vehicles, drones, medical devices and applications (e.g., remote surgery), industrial devices and applications (e.g., robots and / or other wireless devices operating in industrial and / or automated processing chain environments), consumer electronics devices, devices operating on commercial and / or industrial wireless networks, etc. In the following description, the terms "terminal equipment", "communication equipment", "terminal", "user equipment" and "UE" are used interchangeably.

[0038] With the development of communication technology, 3GPP Release 18 has approved the study of the benefits of enhancing the air interface through the latest advances in the field of AI / ML, and AI / ML positioning accuracy enhancement is considered to be one of the three most representative use cases: (a) Channel State Information (CSI) feedback enhancement (e.g., overhead reduction, accuracy improvement, prediction); (b) Beam management (e.g., beam prediction in the time and / or spatial domains to reduce overhead and latency and improve beam selection accuracy); and (c) improving positioning accuracy in different scenarios, including scenarios with severe NLOS conditions.

[0039] For direct AI / ML localization, channel impulse response (CIR) / power delay distribution (PDP) has been agreed upon as candidate model inputs for AI / ML localization. However, further research is needed to investigate the impact of specifications on new model inputs, considering the trade-offs between model performance, model complexity, and computational complexity.

[0040] In view of the above, some embodiments of this disclosure propose a solution for improving AI / ML localization, for example, balancing localization accuracy with model training and storage costs. In some example embodiments of this disclosure, the network device trains a neural network (NN) module based on a pre-trained model and according to the number and combination of TRPs selected by the terminal device for direct AI / ML localization. Therefore, the trained NN module is specific to a particular TRP pattern (both the number and combination) indicated by the terminal device, and the terminal device can perform AI / ML localization based on the downloaded or received trained NN module cascaded after the pre-trained model. That is, for a particular TRP pattern indicated by the terminal device, the network device trains only one lightweight, customized NN model, and the AI / ML localization accuracy meets the accuracy requirements even with a small number of selected TRPs for a particular TRP pattern, thus achieving a trade-off between model performance, model complexity, and computational complexity.

[0041] For illustrative purposes, the following will refer to Figures 1-9 This disclosure describes the principles and exemplary embodiments of an ML framework for enhanced localization. However, it should be noted that these embodiments are given to enable those skilled in the art to understand the inventive concept of this disclosure and to implement the solutions presented herein, and not to limit the scope of this disclosure in any way.

[0042] Figure 1 An example network environment 100 in which exemplary embodiments of the present disclosure may be implemented is illustrated. Environment 100, which may be part of a communication network, includes terminal devices and network devices. Figure 1 As shown, the communication network 100 may include terminal device 110 (hereinafter also referred to as user equipment 110 or UE 110). The communication network 100 may also include network device 120, network device 130 (hereinafter also referred to as transmitter / receiver point (TRP) or site), and network device 140 (hereinafter also referred to as access network device including location management function entity (LMF)). In DL positioning, network device or TRPs 120 and 130 may send a location reference signal (PRS) to terminal device 110, and network device 140 may request UE 110 to collect PRS samples for each selected TRP 120 or 130.

[0043] It should be understood that the number of network devices and terminal devices is for illustrative purposes only and does not represent any limitation. System 100 may include any suitable number of network devices and / or terminal devices suitable for implementing embodiments of this disclosure. Although not shown, it should be understood that one or more terminal devices may be located in environment 100.

[0044] Communication in network environment 100 may be implemented according to any suitable communication protocol(s), including but not limited to third-generation (3G), fourth-generation (4G), fifth-generation (5G) or higher, wireless local area network communication protocols such as IEEE 802.11, and / or any other protocols currently known or to be developed in the future. Furthermore, communication may utilize any suitable wireless communication technology, including but not limited to: multiple-input multiple-output (MIMO), orthogonal frequency division multiplexing (OFDM), time division multiplexing (TDM), frequency division multiplexing (FDM), code division multiplexing (CDM), Bluetooth, ZigBee and machine-type communication (MTC), enhanced mobile broadband (eMBB), massive machine-type communication (mMTC), ultra-reliable low-latency communication (URLLC), carrier aggregation (CA), dual connectivity (DC), and new radio unlicensed (NR-U) technologies.

[0045] It should be understood that Figure 1 The number of devices, their connections, and types shown are for illustrative purposes only and do not represent any limitation. Communication system 100 may include any suitable number of devices appropriate for implementing embodiments of this disclosure.

[0046] In the 3GPP RAN1#112 discussion, it was agreed that CIR / PDP could be used as candidate model inputs for AI / ML localization. It was also agreed to study the impact of the specification on the new model inputs, considering the trade-offs between model performance, model complexity, and computational complexity. At the latest 3GPP RAN1#112-e meeting, it was agreed to evaluate the performance, overhead, and complexity of different approaches with varying model input sizes. To evaluate AI / ML-based localization, studies of model inputs with different TRP numbers included the following methods. Advocates for each method analyzed model performance, signaling overhead (including training data collection and model inference), model complexity, and computational complexity.

[0047] The model input size remains constant. NTRP Method 1, with a value of 18, provides the measured TRP (Temporal Reference Point) for model input. N'TRP The number will vary. When N'TRP < NTRP At that time, the remaining ( NTRP - N'TRP One TRP does not provide measurements to the model input; that is, the measurement value is set to 0. As for Method 1-A: the set of TRPs that provide measurements ( N'TRP The TRP set is fixed. As for method 1-B: provide the measured TRP set ( N'TRP This can change dynamically. It should be noted that for Method 1, a model is provided to cover the entire evaluation area.

[0048] The TRP size of the model input is equal to the TRP used as the model input to provide the measurement. N'TRP Method 2 for the number of ) when N'TRP < NTRP When, given a model, the remainder is ignored ( NTRP - N'TRP ) TRPs. For a given AI / ML model, the measured TRPs are provided ( N'TRP The set is fixed. Regarding method 2-A: a model can be provided to cover the entire evaluation region, which is equivalent to ignoring the remaining ( 18-N'TRP In the case of potential inference of N'TRPs, deploy N'TRPs in the evaluation area for positioning. As for method 2-B, if provided Nmodel ( Nmodel>1 If ) models are used to cover the entire evaluation region, then the total complexity (model complexity is Nmodel (The sum of all models).

[0049] The inventors of this disclosure have evaluated the two methods described above, namely Method 1 and Method 2. For Method 1, in the evaluation, the model input parameter size is fixed for all cases. NTRP =18. When considering a subset of 18 TRPs, the remaining unselected TRPs are set to zero. The positioning accuracy and model complexity are shown in Table 1 below. However, as shown in Table 1, the positioning accuracy is very poor when the number of TRPs in the input CIR / PDP signal is small (e.g., 4 TRPs, 8 TRPs, 12 TRPs). Table 1: Evaluation of gaps for different numbers of TRPs (N_TRP) considering arbitrary TRP selection for Method 1. It is compared with its corresponding model complexity and computational complexity. The input parameters are PDP, N't=128, and UE distribution area=120× 60m。

[0050] As shown in Table 1, when the number of TRPs input to the CIR / PDP signal is small (e.g., 4 TRPs, 8 TRPs, 12 TRPs), the positioning accuracy is quite low. For example, when using 4 TRPs, the horizontal 2D error (measured in meters) at 90% of the CDF is approximately 57.7 m; when using 8 TRPs, the horizontal 2D error (measured in meters) at 90% of the CDF is approximately 32.84 m; and when using 12 TRPs, the horizontal 2D error (measured in meters) at 90% of the CDF is approximately 21.06 m. Clearly, the positioning accuracy does not meet the required accuracy, such as approximately 1 meter.

[0051] For method 2, in the evaluation, the inventors have trained [a program / system] with [specific characteristics]. NMany specific models for a specific TRP pattern. As shown in Table 2, when the number of TRPs is small, the localization accuracy is much higher than that of Method 1. However, in order to adapt to different TRP patterns in real-world implementations, training and storage are necessary. ( ) models, of which N This is the number of TRPs in the TRP mode, which will lead to a huge waste of computing power and memory storage. Table 2: For Method 2—considering arbitrary TRP selection, evaluation is performed on different numbers of TRPs (N_TRP), and It is compared with its corresponding model complexity and computational complexity. The input parameters are PDP, N't=128, and UE distribution area=120× 60m。

[0052] As shown in Table 2, when the number of TRPs in the input CIR / PDP signal is, for example, 4 TRPs, 8 TRPs, and 12 TRPs, the positioning accuracies are 2.3317m, 2.5684m, and 1.3404m, respectively. However, as mentioned above, for signals with... N Each specific TRP pattern of a TRP should be trained and stored. ( This would lead to a huge waste of computing resources and memory storage.

[0053] Based on the above observations and the evaluation of Method 1, when a general TRP pattern model is deployed at the UE, the positioning accuracy may be very poor when only a small subset of TRPs are connected to the UE. Referring to the evaluation of Method 2, acceptable positioning accuracy can be achieved when a specific TRP pattern model is deployed at the UE, but this model is only applicable to specific TRP patterns (the number and combination of TRPs); however, when the TRP pattern (the number or combination of TRPs) changes, a new specific TRP pattern model needs to be trained and stored, which consumes significant computing power and memory.

[0054] To balance localization accuracy and resource consumption (computing power and memory storage, such as computational resources and storage consumption for model training), some embodiments of this disclosure propose a solution for improving artificial intelligence (AI) or machine learning (ML) localization, particularly for deploying TRP-related lightweight NN modules for direct AI / ML localization based on a pre-trained TRP-agnostic general base model (TAB model).

[0055] In the following text, reference will be made to Figure 2 Example signaling procedure 200 is described for performing AI / ML localization. For discussion purposes, please refer to... Figure 1 Describe process 200. Process 200 may involve, for example, Figure 1 The terminal device 110 and network device 140 are shown. It should be understood that, although in Figure 1The process 200 is described in the communication environment 100, but the same process can be applied to other communication scenarios with similar problems.

[0056] like Figure 2 As shown, in process 200, terminal device 110 sends (205) a request to network device 140 for a neural network (NN) module cascaded after the pre-trained model. Network device 140 receives (210) the request. Network device 140 may be a location management function (LMF) entity. Regarding the pre-trained model, reference will be made below. Figure 4 A detailed description is provided. In some embodiments, the pre-trained model is a pre-trained TRP-agnostic general base model (TAB model) for direct AI / ML localization. Terminal device 110 maintains the pre-trained TAB model, and this TAB model may be pre-trained by network device 140 (e.g., LMF). Before sending a request, terminal device 110 may query network device 140 whether it also maintains the same TAB model. If network device 140 maintains the TAB model, network device 140 sends an acknowledgment message to terminal device 110, and then terminal device 110 sends the request to network device 140.

[0057] like Figure 2 As shown, in process 200, network device 140 trains (215) the NN module based on data from a set of Transmitter Receive Points (TRPs) selected by terminal device 110. The TRP set (e.g., TRPs 120 and 130) is selected by terminal device 110, and terminal device 110 may send the indices of the selected TRPs to network device 140, which can then use the data from the selected TRPs to train the NN module. In some embodiments, the trained NN module is a lightweight, custom NN module cascaded after a TAB model. References will be made below. Figure 5 Describe the training of the NN module in detail.

[0058] like Figure 2As shown, in process 200, network device 140 sends (220) the trained NN module cascaded after the pre-trained model and the order of the input signal set from the TRP set to be input to terminal device 110. Terminal device 110 receives (225) the trained NN module and the above order. When input into a pre-trained model (such as a TAB model) held at terminal device 110, the input signal set from the selected TRP set should follow this order. The order of the input signal set to be input to terminal device 110 can be the same as the order of the data from the TRP set to be used by network device 140 to train the NN module. That is, when training the NN module at network device 140, data from the TRP set is input into the model in a specific order, and when the input signal set is to be input to terminal device 110, they can be input in that specific order.

[0059] like Figure 2 As shown, in process 200, terminal device 110 performs (230) artificial intelligence (AI) / machine learning (ML) localization using a set of input signals from the TRP set in the order described above, based on a trained NN module cascaded after a pre-trained model. In some embodiments, the input signals from the TRP set may include channel impulse response (CIR) signals or power delay distribution (PDP) signals, and each input signal may have an index of the corresponding TRP.

[0060] In process 200 of this disclosure, network device 140 trains a lightweight neural network module based on a pre-trained model (such as a TAB model) according to the number and combination of TRPs selected by terminal device 110 for direct AI / ML localization. That is, the trained neural network module is specific to a particular TRP pattern (number and combination of TRPs) indicated by terminal device 110. Terminal device 100 then downloads and deploys the trained neural network module at the output of the pre-trained model; that is, the downloaded trained neural network module is cascaded after the pre-trained model. Terminal device 100 can then perform AI / ML localization based on the trained neural network module cascaded after the pre-trained model. In other words, for a particular TRP pattern with N selected TRPs indicated by the terminal device, the network device trains and stores only one lightweight, customized neural network model for that particular TRP pattern, instead of according to method 2 described above. This significantly reduces training complexity, computational power, and storage memory. Furthermore, for each TRP mode, when direct AI / ML localization is performed, the trained NN module is cascaded after the pre-trained general model. Both the trained NN module and the trained general model contribute to the calculation of localization coordinates, and therefore, even with several selected TRPs for a particular TRP mode, the AI / ML localization accuracy will meet the accuracy requirements and will be higher than the accuracy obtained by Method 1 described above.

[0061] In the following text, reference will be made to Figure 3 This document describes an example signaling process according to some embodiments of the present disclosure for deploying a custom neural network (NN) module cascaded after a frozen TAB model on the terminal device side for direct AI / ML localization.

[0062] like Figure 3 As shown, at step 0, UE 110 should maintain the TAB model, which is a prerequisite for this framework. For example, the network device (e.g., LMF 140) can pre-train this TAB model. At step 1, UE 110 queries, for example, via a message whether LMF 140 maintains the same TAB model. If LMF 140 does not maintain the same TAB model, the process terminates.

[0063] If LMF 140 maintains the same TAB model, the process continues. At step 2, LMF 140 sends a confirmation of maintaining the same TAB model to UE 110 via a new message. Upon receiving the confirmation of maintaining the same TAB model from LMF 140, at step 3, UE 110 sends a request via a new message to LMF 140 for a lightweight, customized NN module cascaded after the TAB model for direct AI / ML localization. Figure 3 As shown, at step 4, UE 110 indicates the indexes of its selected TRPs 120 and 130 to LMF 140. For example, the index of the selected TRP is #9#15#18.

[0064] At step 5, LMF 140 uses data from the selected subset of TRPs (e.g., TRPs 120 and 130) to train a TRP-related lightweight neural network (NN) module cascaded after the TAB model for direct AI / ML localization. It should be noted that during training, the TAB model is frozen, and its parameters are not tuned; only the parameters in the lightweight custom NN module are tuned. CIR / PDP signals from the selected subset of TRPs are fed into the TAB model in a configured TRP index order. For example, the configured TRP index order is #15#18#9, rather than the indices of the selected TRPs as indicated by UE 110 (e.g., #9#15#18).

[0065] In step 6, LMF 140 sends the trained TRP-related lightweight NN module to UE 110. For example... Figure 3 As shown, at step 7, LMF 140 indicates to UE 110 the TRP index order of the input signals (e.g., #15#18#9) for use by UE 110 when feeding the signals into the TRP-related lightweight NN module cascaded after the TAB model at UE 110 for direct AI / ML localization. For example, if the selected TRP index is #9#15#18, and the TRP index order of the input signals is configured by LMF 140 as #15#18#9, then UE 110 should input these signals into the model in the order #15#18#9 (instead of #9#15#18 or some other index order).

[0066] At step 8, LMF 140 indicates the selected TRP index (e.g., #9#15#18) to the selected TRP. Figure 3 As shown, at step 9, the selected TRP sends its CIR / PDP signal and TRP index to UE 110. At step 10, UE 110 performs AI / ML localization using the received CIR / PDP signals from the selected TRP in the indicated input order (e.g., order #15#18#9) using the TAB model of the cascaded TRP-related lightweight NN module. It should be noted that the number of TRPs 120 and 130 is not limited to two TRPs, and TRPs 120 and 130 are described for illustrative purposes only.

[0067] Therefore, in some embodiments of this disclosure, a solution is proposed for deploying TRP-related lightweight NN modules on the UE side based on a pre-trained TRP-agnostic general base model (TAB model) for direct AI / ML localization. In this solution, the NW (e.g., LMF 140) pre-trains the TAB model and then trains the TRP-related lightweight NN modules according to the UE's request in dynamic TRP scenarios. The UE 110 also maintains the same pre-trained TAB model as a prerequisite and then requests and downloads a TAB-compatible lightweight custom NN module from the LMF 140, which can be immediately cascaded after the TAB model. Since the TAB model is a general TRP-agnostic pre-trained model, when the UE 110 receives CIR / PDP signals from any TRP mode (the number of TRPs is denoted as M), these signals will pass through the TAB model to generate M intermediate outputs. These M intermediate outputs are then fed into the cascaded custom NN module to generate localization results, such as two-dimensional localization coordinates. Here, by merging M intermediate features from a specific TRP pattern, a cascaded custom NN module learns TRP-related localization.

[0068] To evaluate the localization accuracy of a lightweight custom NN module cascaded immediately after the TAB model, the inventors of this disclosure evaluated two schemes for direct AI / ML localization using datasets from TRP#9, #15, and #18. It should be noted that the datasets from TRP#8, #15, and #18 were not included in the training dataset and were used only to test the prediction accuracy of the trained model.

[0069] In the first approach, according to some embodiments of this disclosure, the evaluation is performed based on the proposed TAB model and a lightweight custom NN module cascaded after the TAB model. In the second approach, the evaluation is performed based on the baseline ViT, which is a specific TRP pattern model trained according to the conventional method 2 described above, with 196,646 data samples used for training and the remaining data samples used for testing.

[0070] Table 3 below compares the trainable parameters and localization accuracy at CDF90 between the first and second schemes. As shown in Table 3, compared with the baseline ViT scheme according to conventional method 2, the scheme proposed according to some embodiments of this disclosure can achieve a localization accuracy of 1.40m at CDF90, while reducing the training complexity by 99.86%, that is, the trainable parameters are reduced from 42M in the second scheme to about 60K in the first scheme. Table 3. Performance Comparison of Direct Location Tasks

[0071] As can be seen from the comparison in Table 3, the positioning accuracy of 1.40m is close to the accuracy of 0.826m obtained by the traditional method. In other words, the model performance is qualified, while the model training and computational complexity are significantly reduced by 99.86%.

[0072] In some embodiments of this disclosure, based on a pre-trained TAB model, NW 140 trains a lightweight custom NN module for direct AI / ML localization according to the number and combination of TRPs connected to the UE, thereby facilitating the customization of this TRP-related NN module based on the TAB model. In this process, UE 110 and LMF 140 are aligned, and both are ensured to have the same TAB model. If it is verified that UE 110 and LMF 140 share the same TAB model, UE 110 specifically sends a request to LMF 140 for a lightweight custom NN module cascaded after the TAB model for direct AI / ML localization based on a selected TRP pattern. LMF 140 then uses data from a selected subset of TRPs to train the lightweight custom NN module cascaded after the frozen TAB model for direct AI / ML localization. LMF 140 sends the trained lightweight custom NN module to UE 110 and indicates the TRP index order of the input signals used for model inference. Then, the UE 110 downloads and deploys a trained lightweight custom NN module at the output of the TAB model for TRP-related AI / ML direct localization. The UE 110 then inputs the CIR / PDP signals from the selected TRP into the cascaded custom NN module's TAB model in the indicated input order for direct AI / ML localization. Therefore, for models with... N A specific TRP pattern for a selected TRP, with only one lightweight custom NN module (rather than) A TAB model is trained and stored, and both the TAB model and the trained lightweight NN module cascaded after the TAB model can contribute to obtaining high-precision localization results.

[0073] In the following text, reference will be made to Figure 4 Example implementations for training general models, such as the TAB model, are described according to some embodiments of this disclosure.

[0074] like Figure 4As shown, the TAB model is an image-text cross-modal pre-trained model, which consists of two key feature extractors / encoders. The first key feature extractor or encoder is associated with the contextual semantic embedding block <text encoder>: a description of the physical environment (e.g., NLOS / NLOS classification, BS (TRP) location, UE location, environment classification, etc.) is input in the form of natural language text. The second key feature extractor or encoder is associated with the wireless channel feature extraction block <image encoder>: wireless channel features (e.g., CSI, CIR, PDP, etc.) are input in the form of an image.

[0075] By maximizing the similarity between the output matrices from different models, the text encoder and image encoder are jointly trained to align multimodal information in a shared embedding space, thereby achieving cross-modal semantic understanding and matching. Therefore, the text-image cross-modal foundational model learns inherent wireless environment knowledge and becomes a general model that is TRP mode-agnostic.

[0076] It should be noted that text encoding will be used when training the TAB model, but not when training the lightweight NN module or performing AI / ML localization. That is, when training the lightweight NN module, data from the selected TRP can be fed into the image encoder of the TAB model to generate the corresponding intermediate output, and this intermediate output can be fed into the NN module to generate 2D localization coordinates. Therefore, a text encoder is not used when training the NN module. A similar process will be performed when performing AI / ML localization, and a text encoder will not be used.

[0077] In the following text, reference will be made to Figure 5 This document describes example implementations of lightweight custom neural network (NN) modules cascaded after a frozen TAB model, according to some embodiments of the present disclosure.

[0078] Since the pre-trained TAB model learns useful representations and semantic relationships between images and text during the pre-training phase, it provides a strong initial starting point. Therefore, building upon the ability to inherit and build upon the pre-trained TAB model, a lightweight custom NN module was trained to align with direct AI / ML localization tasks.

[0079] like Figure 5As shown, a lightweight custom neural network (NN) module is cascaded after the image encoder of the TAB model. During training of the cascaded NN module, the image encoder is frozen, so that the parameters of the image encoder in the pre-trained TAB model are not tuned, and only the parameters of the lightweight custom NN module are tuned. The lightweight custom NN module is trained using data from selected TRPs (such as TRP#1#2#3) in a specific input order (e.g., TRP#2#3#1). This means that the input to the image encoder is a set of samples comprising multiple CSI samples from selected TRPs in a specific order (e.g., TRP#2#3#1). By inputting the CSI sample sets with the specific input order into the image encoder, the output vectors from the same sample sets are cascaded in the same specific order and used as one input to the lightweight custom NN module to obtain the corresponding localization coordinates.

[0080] In one example, the number of TRPs selected is 3, and the internal architecture of the lightweight custom NN module is shown in Table 4. Table 4: Internal Architecture of Lightweight Custom NN Module

[0081] As shown in Table 4, the input size N1 The size is 384, and the final output size is a two-dimensional positioning coordinate. Therefore, a lightweight custom NN module can be used to predict positioning results, such as two-dimensional positioning coordinates.

[0082] Figure 6 A flowchart illustrating an example method 600 implemented at a terminal device according to some other embodiments of the present disclosure is shown. Reference will be made to this flowchart for discussion purposes. Figure 1 Method 600 is described from the perspective of terminal device 110.

[0083] At box 610, terminal device 110 sends a request to network device 140 for a neural network module cascaded after the pre-trained model. At box 620, terminal device 100 receives from network device 140 the trained neural network module cascaded after the pre-trained model and the input signal set from the Transmitter Receiver Point (TRP) set to be input to the terminal device in the order described above. At box 630, terminal device 100 performs artificial intelligence (AI) / machine learning (ML) localization based on the trained neural network module cascaded after the pre-trained model, using the input signal set from the TRP set in the order described above.

[0084] In some embodiments, the terminal device 110 also maintains a pre-trained model; and determines whether the network device maintains the same pre-trained model. In some embodiments, the pre-trained model is a TRP-agnostic general basis (TAB) model and includes an image encoder; and the trained neural network module is a TRP-specific, lightweight trained neural network module cascaded after the image encoder.

[0085] In some embodiments, terminal device 110 also sends an index of a TRP set selected by terminal device 140 to network device 140, wherein the neural network module is trained by network device 140 based on data from the TRP set. In some embodiments, terminal device 110 also receives a set of input signals from the TRP set, each input signal having an index of a corresponding TRP.

[0086] In some embodiments, the terminal device 110 performs AI / ML localization by passing the input signal set from the TRP set through a pre-trained model in the order described above to generate an intermediate output set; and by inputting the intermediate output set into a trained neural network module to generate a localization result.

[0087] In some embodiments, the input signals include at least one of a channel impulse response (CIR) signal or a power delay distribution (PDP) signal. In some embodiments, the order of the input signal set is the same as the order of the data from the TRP set used to train the neural network module. In some embodiments, the network device includes a location management function entity.

[0088] Figure 7 A flowchart illustrating an example method 700 implemented at a network device according to some other embodiments of the present disclosure is shown. Reference will be made to this flowchart for discussion purposes. Figure 1 Method 700 is described from the perspective of network device 140.

[0089] At block 710, network device 140 receives a request from terminal device 110 for a neural network module cascaded after a pre-trained model. At block 720, network device 140 trains the neural network module based on data from a set of Transmitter Receiver Points (TRPs) selected by the terminal device. At block 730, network device 140 sends to terminal device 110 the trained neural network module cascaded after the pre-trained model, along with a set of input signals from the TRP set, which are to be fed into the pre-trained model cascaded with the trained neural network module at the terminal device.

[0090] In some embodiments, network device 140 also receives a signal from terminal device 110 for querying whether the network device maintains the pre-trained model; and sends the query result to terminal device 110.

[0091] In some embodiments, the pre-trained model is a TRP-agnostic generalized foundation (TAB) model and includes an image encoder; and the trained neural network module is a TRP-specific, lightweight neural network module cascaded after the image encoder. In some embodiments, the pre-trained model also includes a text encoder.

[0092] In some embodiments, network device 140 also receives an index of the TRP set from terminal device 110. In some embodiments, network device 140 trains a neural network module by: passing data from the TRP set through an image encoder in the order described above to generate an intermediate output set; generating a localization result by inputting the intermediate output set into the neural network module; and tuning at least one parameter of the neural network module based on the generated localization result to obtain a trained neural network module. In some embodiments, network device 140 includes a location management function entity.

[0093] In some embodiments, an apparatus capable of performing method 600 (e.g., terminal device 110) may include components for performing corresponding steps of method 600. These components may be implemented in any suitable form. For example, the components may be implemented in a circuit system or a software module.

[0094] In some embodiments, the apparatus includes: components for sending a request to a network device for a neural network module cascaded after a pre-trained model; components for receiving from the network device the trained neural network module cascaded after the pre-trained model and the sequence of input signal sets from a set of transmit receiving points (TRPs) to be input to a terminal device; and components for performing artificial intelligence (AI) / machine learning (ML) localization based on the trained neural network module cascaded after the pre-trained model using the input signal sets from the TRP set in the aforementioned sequence.

[0095] In some embodiments, the apparatus further includes components for maintaining a pre-trained model and for determining whether the network device maintains the same pre-trained model.

[0096] In some embodiments, the apparatus further includes components for the network device to transmit an index of a TRP set selected by the terminal device, wherein the neural network module is trained by the network device based on data from the TRP set. In some embodiments, the apparatus further includes components for receiving a set of input signals from the TRP set, each input signal having an index of a corresponding TRP.

[0097] In some embodiments, the apparatus further includes components for performing additional steps of some embodiments of method 600. In some embodiments, the components include at least one processor and at least one memory including computer program code, the at least one memory and the computer program code being configured to, together with the at least one processor, implement the performance of the apparatus.

[0098] In some embodiments, an apparatus capable of performing method 700 (e.g., network device 140) may include components for performing the corresponding steps of method 700. These components may be implemented in any suitable form. For example, the components may be implemented in a circuit system or a software module.

[0099] In some embodiments, the apparatus includes components for receiving from a terminal device a request for a neural network module cascaded after a pre-trained model; components for training the neural network module based on data from a set of Transmitter Receiver Points (TRPs) selected by the terminal device; and components for sending to the terminal device the trained neural network module cascaded after the pre-trained model and the sequence of an input signal set from the TRP set, wherein the input signal set from the TRP set is to be input into the pre-trained model cascaded with the trained neural network module at the terminal device.

[0100] In some embodiments, the apparatus further includes: a component for receiving from a terminal device a signal for querying whether a network device maintains a pre-trained model; and a component for sending the query result to the terminal device.

[0101] In some embodiments, the apparatus further includes components for receiving an index of the TRP set from a terminal device. In some embodiments, the components for training the neural network module train the neural network by: passing data from the TRP set through an image encoder in the order described above to generate an intermediate output set; generating a localization result by inputting the intermediate output set into the neural network module; and tuning at least one parameter of the neural network module based on the generated localization result to obtain a trained neural network module.

[0102] In some embodiments, the apparatus further includes components for performing additional steps of some embodiments of method 700. In some embodiments, the components include at least one processor and at least one memory including computer program code, the at least one memory and the computer program code being configured to, together with the at least one processor, implement the performance of the apparatus.

[0103] Figure 8 This is a simplified block diagram of a device 800 suitable for implementing embodiments of the present disclosure. The device 800 can be provided to implement a communication device, such as... Figure 1The terminal device 110 and network device 140 are shown. As shown, device 800 includes one or more processors 810, one or more memories 820 coupled to processor 810, and one or more communication modules 840 coupled to processor 810.

[0104] The communication module 840 is used for bidirectional communication. The communication module 840 has at least one antenna to facilitate communication. The communication interface can represent any interface required for communication with other network devices.

[0105] Processor 810 can be of any type suitable for a local technology network, and by way of non-limiting example, can include one or more of the following: general-purpose computer, special-purpose computer, microprocessor, digital signal processor (DSP), and processor based on a multi-core processor architecture. Device 800 can have multiple processors, such as application-specific integrated circuit chips that are time-dependent on a clock synchronized with the main processor.

[0106] Memory 820 may include one or more non-volatile memories and one or more volatile memories. Examples of non-volatile memories include, but are not limited to, read-only memory (ROM) 824, electrically programmable read-only memory (EPROM), flash memory, hard disk, compact disc (CD), digital video disc (DVD), and other magnetic and / or optical storage devices. Examples of volatile memories include, but are not limited to, random access memory (RAM) 822 and other volatile memories that do not persist during power outages.

[0107] Computer program 830 includes computer-executable instructions that are executed by the associated processor 810. Program 830 may be stored in ROM 824. Processor 810 may perform any suitable actions and processes by loading program 830 into RAM 822.

[0108] The embodiments of this disclosure can be implemented by a program, such that device 800 can execute the reference. Figure 6 and Figure 7 Any process discussed in this disclosure. Embodiments of this disclosure may also be implemented by hardware or by a combination of software and hardware.

[0109] In some embodiments, program 830 may be tangibly contained in a computer-readable medium, which may be included in device 800 (such as memory 820) or other storage device accessible to device 800. Device 800 may load program 830 from the computer-readable medium into RAM 822 for execution. The computer-readable medium may include any type of tangible non-volatile memory, such as ROM, EPROM, flash memory, hard disk, CD, DVD, etc.

[0110] Figure 9 An example of a computer-readable medium 900 in the form of a CD or DVD according to some embodiments of the present disclosure is illustrated. A program 930 is stored on the computer-readable medium. It should be noted that although the computer-readable medium 900 is depicted in the form of a CD or DVD, the computer-readable medium 900 may be any other form suitable for carrying or storing the program 930.

[0111] Generally, the various embodiments of this disclosure can be implemented using hardware or dedicated circuitry, software, logic, or any combination thereof. Some aspects can be implemented using hardware, while others can be implemented using firmware or software that can be executed by a controller, microprocessor, or other computing device. Although various aspects of the embodiments of this disclosure are illustrated and described as block diagrams, flowcharts, or using some other graphical representation, it should be understood that, as non-limiting examples, the blocks, apparatuses, systems, techniques, or methods described herein can be implemented using hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.

[0112] This 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 instructions included in a program module, which execute in a device on a target real or virtual processor to perform the above-mentioned... Figures 6 to 7 The method described is 600 or 700. Typically, a program module includes routines, programs, libraries, objects, classes, components, data structures, etc., that perform specific tasks or implement specific abstract data types. In various embodiments, the functionality of a program module can be combined or split among program modules as needed. The machine-executable instructions of a program module can be executed on a local or distributed device. In a distributed device, a program module can reside on both local and remote storage media.

[0113] Program code used to perform the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a stand-alone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0114] In the context of this disclosure, computer program code or related data may be carried by any suitable carrier to enable a device, apparatus, or processor to perform the various processes and operations described above. Examples of carriers include signals, computer-readable media, etc.

[0115] Computer-readable media can be computer-readable signal media or computer-readable storage media. Computer-readable media can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any suitable combination of the foregoing. More specific examples of computer-readable storage media will include electrical connections having one or more wires, portable computer floppy disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable optical disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing. The term "non-transient" as used herein is a limitation on the medium itself (i.e., tangible, not signaling), not a limitation on the persistence of data storage (e.g., RAM and ROM).

[0116] Furthermore, although operations are described in a specific order, this should not be construed as requiring the operations to be performed in the specific order shown or in sequential order, or to perform all of the shown operations to obtain the desired result. In some cases, multitasking and parallel processing may be advantageous. Similarly, while several specific implementation details are included in the foregoing discussion, these should not be construed as limiting the scope of this disclosure, but rather as descriptions of features that may be specific to particular embodiments. Certain features described in the context of a single embodiment may also be implemented in combination in a single embodiment. Conversely, the various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.

[0117] Although this disclosure has been described in language specific to structural features and / or methodological actions, it should be understood that the disclosure as defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features or actions described above are disclosed as exemplary forms of implementing the claims.

Claims

1. A terminal device, comprising: At least one processor; as well as At least one memory stores instructions that, when executed by the at least one processor, cause the terminal device to at least: Send a request to the network device for the neural network module cascaded after the pre-trained model; The order in which the network device receives the trained neural network module cascaded after the pre-trained model and the set of input signals from the Transmitter Receiver Point (TRP) set to be input to the terminal device; as well as Artificial intelligence (AI) / machine learning (ML) localization is performed by using the set of input signals from the TRP set in the order described, based on the trained neural network module cascaded after the pre-trained model.

2. The terminal device according to claim 1, wherein the terminal device is further configured to: Maintain the pre-trained model; and Determine whether the network devices maintain the same pre-trained model.

3. The terminal device according to claim 1 or 2, wherein: The pre-trained model is a TRP-agnostic universal foundation (TAB) model and includes an image encoder; and The trained neural network module is a TRP-specific, lightweight neural network module cascaded after the image encoder.

4. The terminal device according to any one of claims 1 to 3, wherein the terminal device is further configured to: The network device sends an index of the TRP set selected by the terminal device, wherein the neural network module is trained by the network device based on data from the TRP set.

5. The terminal device according to any one of claims 1 to 4, wherein the terminal device is further configured to: The input signal set is received from the TRP set, each input signal having an index of a corresponding TRP.

6. The terminal device according to any one of claims 1 to 5, wherein the terminal device is configured to perform the AI / ML positioning in the following manner: The input signal set from the TRP set is passed through the pre-trained model in the stated order to generate an intermediate output set; and The localization result is generated by inputting the intermediate output set into the trained neural network module.

7. The terminal device according to any one of claims 1 to 6, wherein the input signal includes at least one of a channel impulse response (CIR) signal or a power delay distribution (PDP) signal.

8. The terminal device according to any one of claims 4 to 7, wherein the order of the input signal set is the same as the order of the data from the TRP set used to train the neural network module.

9. The terminal device according to any one of claims 1 to 8, wherein the network device includes a location management function entity.

10. A network device, comprising: At least one processor; as well as At least one memory storing instructions that, when executed by the at least one processor, cause the network device to at least: Receive requests from the terminal device for neural network modules cascaded after the pre-trained model; The neural network module is trained based on data from the Transmitter Receiver Point (TRP) set selected by the terminal device. as well as The training neural network module cascaded after the pre-trained model and the input signal set from the TRP set are sent to the terminal device in the following order, wherein the input signal set from the TRP set is to be input into the pre-trained model cascaded with the training neural network module at the terminal device.

11. The network device of claim 10, wherein the network device is further configured to: Receive from the terminal device a signal for querying whether the network device maintains the pre-trained model; and The query result is sent to the terminal device.

12. The network device according to claim 10 or 11, wherein: The pre-trained model is a TRP-agnostic universal foundation (TAB) model and includes an image encoder; and The trained neural network module is a TRP-specific, lightweight neural network module cascaded after the image encoder.

13. The network device of claim 12, wherein the pre-trained model further comprises a text encoder.

14. The network device according to any one of claims 10 to 13, wherein the network device is further configured to: Receive the index of the TRP set from the terminal device.

15. The network device of claim 12, wherein the network device is configured to train the neural network module in the following manner: Data from the TRP set is passed through the image encoder in the stated order to generate an intermediate output set; The localization result is generated by inputting the intermediate output set into the neural network module; as well as Based on the generated localization results, at least one parameter of the neural network module is tuned to obtain the trained neural network module.

16. The network device according to any one of claims 10 to 15, wherein the network device includes a location management function entity.

17. A method comprising: Send a request to the network device for the neural network module cascaded after the pre-trained model; The order in which the network device receives the trained neural network module cascaded after the pre-trained model and the set of input signals from the Transmitter Receiver Point (TRP) set to be input to the terminal device; as well as Artificial intelligence (AI) / machine learning (ML) localization is performed by using the set of input signals from the TRP set in the order described, based on the trained neural network module cascaded after the pre-trained model.

18. A method comprising: Receive requests from the terminal device for neural network modules cascaded after the pre-trained model; The neural network module is trained based on data from the Transmitter Receiver Point (TRP) set selected by the terminal device. as well as The training neural network module cascaded after the pre-trained model and the input signal set from the TRP set are sent to the terminal device in the following order, wherein the input signal set from the TRP set is to be input into the pre-trained model cascaded with the training neural network module at the terminal device.

19. An apparatus comprising: A component used to send requests to network devices for neural network modules cascaded after a pre-trained model; A component for receiving, from the network device, a trained neural network module cascaded after the pre-trained model and a set of input signals from the Transmitter Receiver Point (TRP) set to be input to the terminal device in sequence; as well as A component for performing artificial intelligence (AI) / machine learning (ML) localization based on the trained neural network module cascaded after the pre-trained model by using the set of input signals from the TRP set in the order stated.

20. An apparatus comprising: A component for receiving requests from a terminal device for neural network modules cascaded after a pre-trained model; Components for training the neural network module based on data from a set of Transmitter Receiver Points (TRPs) selected by the terminal device; as well as A component for sending to the terminal device a sequence of trained neural network modules cascaded after the pre-trained model and an input signal set from the TRP set, wherein the input signal set from the TRP set is to be input into the pre-trained model cascaded with the trained neural network modules at the terminal device.

21. A non-transitory computer-readable medium comprising program instructions that, when executed by a device, cause the device to perform at least the method according to claim 17.

22. A non-transitory computer-readable medium comprising program instructions that, when executed by a device, cause the device to perform at least the method according to claim 18.

Images