Data efficient update for channel classification

Abstract

Example embodiments of the present disclosure relate to data efficient updates for channel classification. The device determines an importance level of channel measurement information related to a communication channel for updating a classification model configured to determine an estimated classification result of the communication channel based on the channel measurement information. If it is determined that the importance level of the channel measurement information exceeds an importance threshold for the classification model, the device obtains a truth classification result for the communication channel. The device causes the classification model to be updated based at least in part on the channel measurement information and the truth classification result.

Description

Cross-references to related applications

[0001] This application claims priority to the international patent application entitled “ON-DEMAND LABELLING FOR CHANNEL CLASSIICATION TRAINING” and application number PCT / CN2022 / 105916 filed on July 15, 2022, the entire contents of which are incorporated by reference into this application. Technical Field

[0002] Embodiments of the present disclosure relate generally to the field of telecommunications, and in particular, to methods, devices, apparatuses, and computer-readable storage media for efficient updating of data for channel classification. Background Art

[0003] Location awareness is a fundamental aspect of wireless communication networks and will enable countless location-enabled services in different applications. As technology develops, the integration and utilization of location information in everyday applications will grow significantly.

[0004] Many positioning techniques that depend on techniques such as time of arrival (TOA), time difference of arrival (TDOA), and angle of arrival (AOA) require line-of-sight (LOS) propagation between a reference point (e.g., network equipment) and the mobile device to be positioned. However, for non-line-of-sight (NLOS) propagation cases in indoor / outdoor environments, positioning accuracy deteriorates significantly due to the inability to identify reflected multipath radio frequency (RF) propagations from different angles of arrival with different delay spreads. On the other hand, artificial intelligence (AI) algorithms are inherently superior in fingerprinting positioning inference in terms of accuracy and efficiency, both in LOS and NLOS conditions. Therefore, it is important to classify channel propagation, at least because of its impact on the choice of positioning method. Summary of the invention

[0005] The scope of protection sought by various embodiments of the present invention is set out by the independent claims. The embodiments / examples and features (if any) described in this specification that do not fall within the scope of the independent claims should be interpreted as examples useful for understanding the various embodiments of the present invention. Please note that the term "embodiment" or "example" should be applied accordingly to the terms used in this application, that is, if the term "example" is used, the sentence should accordingly speak of "example", or if the term "embodiment" is used, the sentence should accordingly speak of "embodiment".

[0006] Embodiments that do not fall within the scope of the claims, if any, are to be construed as examples useful for understanding the various embodiments of the present disclosure.

[0007] In a first aspect, a device is provided. The device includes: at least one processor; and at least one memory storing instructions, which when executed by the at least one processor cause the first device to at least perform: determining an importance level of channel measurement information related to a communication channel for updating a classification model, the classification model being configured to determine an estimated classification result of the communication channel based on the channel measurement information; if it is determined that the importance level of the channel measurement information exceeds an importance threshold for the classification model, obtaining a true classification result for the communication channel; and causing the classification model to be updated based at least in part on the channel measurement information and the true classification result.

[0008] In a second aspect, a method is provided. The method includes: determining an importance level of channel measurement information related to a communication channel for updating a classification model, the classification model being configured to determine an estimated classification result of the communication channel based on the channel measurement information; obtaining a true classification result for the communication channel if it is determined that the importance level of the channel measurement information exceeds an importance threshold for the classification model; and causing the classification model to be updated based at least in part on the channel measurement information and the true classification result.

[0009] In a third aspect, an apparatus is provided. The apparatus includes: a component for determining an importance level of channel measurement information associated with a communication channel for updating a classification model, the classification model being configured to determine an estimated classification result of the communication channel based on the channel measurement information; a component for obtaining a true classification result for the communication channel if the importance level of the channel measurement information is determined to exceed an importance threshold for the classification model; and a component for causing the classification model to be updated based at least in part on the channel measurement information and the true classification result.

[0010] In a fourth aspect, a computer-readable medium is provided, wherein the computer-readable medium includes instructions stored thereon, the instructions being used to cause an apparatus to at least perform the method according to the second aspect.

[0011] It should be understood that the invention summary is not intended to identify the key or essential features of the 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 readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Some example embodiments will now be described with reference to the accompanying drawings, in which:

[0013] Figure 1 shows an example communication environment in which example embodiments of the present disclosure may be implemented;

[0014] Figure 2 A flowchart showing a process for efficient data update of a classification model according to some example embodiments of the present disclosure is shown;

[0015] Figure 3 shows an example signaling flow for communication according to some example embodiments of the present disclosure;

[0016] Figure 4 shows an example signaling flow for communication according to some example embodiments of the present disclosure;

[0017] Figure 5 A flowchart illustrating a process for determining an uncertainty level according to some example embodiments of the present disclosure;

[0018] Fig. 6A Examples of classification models and reference classification models generated therefrom according to some example embodiments of the present disclosure are shown;

[0019] Figure 6B shows examples of classification models and reference classification models generated therefrom according to some further example embodiments of the present disclosure;

[0020] Figure 7 The model performance gains of some example embodiments of the present disclosure relative to traditional model training methods are shown;

[0021] Figure 8 A flowchart illustrating a process for determining an uncertainty level according to some further example embodiments of the present disclosure;

[0022] Fig. 9A Another example of a classification model according to some further example embodiments of the present disclosure is shown;

[0023] Fig. 9B shows further examples of classification models according to some further example embodiments of the present disclosure;

[0024] FIG. 10A to FIG. 10B The model performance gains of some example embodiments of the present disclosure relative to traditional model training methods are shown;

[0025] Fig.11 shows a simplified block diagram of an apparatus suitable for implementing an example embodiment of the present disclosure; and

[0026] Fig.12 A block diagram of an example computer-readable medium is shown according to some example embodiments of the present disclosure.

[0027] Throughout the drawings, the same or similar reference numerals represent the same or similar elements. Throughout the drawings, the same or similar reference numerals represent the same or similar elements. DETAILED DESCRIPTION

[0028] The principles of the present disclosure will now be described with reference to some example embodiments. It should be understood that these embodiments are described only for illustrative purposes and are helpful for those skilled in the art to understand and implement the present disclosure without implying any limitation on the scope of the present disclosure. The embodiments described herein can be implemented in various ways different from the ways 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 skill in the art to which this disclosure belongs.

[0030] References in this disclosure to "one embodiment," "an embodiment," "an example embodiment," etc. indicate that the described embodiment may include a particular feature, structure, or characteristic, but not every embodiment necessarily includes the particular feature, structure, or characteristic. Furthermore, such phrases do not necessarily refer to the same embodiment. Furthermore, when a particular feature, structure, or characteristic is described in conjunction with an embodiment, it is considered that it is within the knowledge of those skilled in the art to affect such feature, structure, or characteristic in conjunction with other embodiments, whether or not explicitly described.

[0031] It should be understood that although the terms "first", "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 element. For example, without departing from the scope of the exemplary embodiments, the first element may be referred to as the second element, and similarly, the second element may be referred to as the first element. As used herein, the term "and / or" includes any and all combinations of one or more of the listed terms.

[0032] As used herein, “at least one of: ” and “at least one of ” and similar expressions, where a list of two or more elements is connected by “and” or “or”, mean at least any one of the elements, or at least any two or more of the elements, or at least all of the elements.

[0033] The terms used herein are only used for the purpose of describing specific embodiments and are not intended to limit the exemplary embodiments. As used herein, the singular forms "a", "an", and "the" are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the terms "include", "comprises", "has", "includes", and / or "comprising" when used herein specify the presence of the features, elements, and / or components, etc., but do not exclude the presence or addition of one or more other features, elements, components, and / or combinations thereof.

[0034] As used in this application, the term "circuitry" may refer to one or more or all of the following: (a) hardware circuit implementation only (such as implementation in analog circuitry and / or digital circuitry only), and (b) a combination of hardware circuitry and software, such as (where applicable): (i) a combination of analog and / or digital hardware circuitry and software / firmware, and (ii) any portion of hardware processor(s) (including digital signal processor(s)) with software, software and memory(s) that work together to enable a device (such as a mobile phone or server) to perform various functions, and (c) Hardware circuits and or processor(s), such as microprocessor(s) or portions of microprocessor(s), that require software (e.g., firmware) for operation, but the software may not be present when not required for operation.

[0035] This definition of circuitry applies to all uses of the term in this application, including any claims. As another example, as used in this application, the term circuitry also covers an implementation of only a hardware circuit, or a processor (or multiple processors), or a portion of a hardware circuit or processor and its (or their) accompanying software and / or firmware. The term circuitry also covers (for example and if applicable to a particular claim element) a baseband integrated circuit or processor integrated circuit for a mobile device or a similar integrated circuit in a server, cellular network device, or other computing or network device.

[0036] As used herein, the term "communication network" refers to a network that follows any suitable communication standard, such as New Radio (NR), Long Term Evolution (LTE), Advanced LTE (LTE-A), Wideband Code Division Multiple Access (WCDMA), High Speed ​​Packet Access (HSPA), Narrowband Internet of Things (NB-IoT), etc. In addition, the communication between the terminal equipment and the network equipment in the communication network can be performed according to any suitable generation communication protocol, including but not limited to the first generation (1G), second generation (2G), 2.5G, 2.75G, third generation (3G), fourth generation (4G), 4.5G, fifth generation (5G) communication protocol and / or any other protocol currently known or developed in the future. The embodiments of the present disclosure can be applied to various communication systems. In view of the rapid development of communication, there will certainly be communication technologies and systems of future types that can embody the present disclosure. It should not be considered that the scope of the present disclosure is limited to the aforementioned system.

[0037] As used herein, the term "network device" refers to a node in a communication network, via which a terminal device accesses the network and receives services from the node. A network device may refer to a base station (BS) or an access point (AP), for example, a NodeB (NodeB or NB), an evolved NodeB (eNodeB or eNB), a NR NB (also referred to as a gNB), a remote radio unit (RRU), a radio head (RH), a remote radio head (RRH), a relay, an integrated access and backhaul (IAB) node, a low-power node, such as a femto, a micro, a non-terrestrial network (NTN) or a non-terrestrial network device, such as a satellite network device, a low earth orbit (LEO) satellite and a geosynchronous earth orbit (GEO) satellite, an aircraft network device, etc., depending on the terminology and technology applied. In some example embodiments, a radio access network (RAN) split architecture includes a centralized unit (CU) and a distributed unit (DU) at an IAB donor node. An IAB node includes a mobile terminal (IAB-MT) portion, which behaves similarly to a UE for a parent node, and the DU portion of the IAB node behaves similarly to a base station for a next-hop IAB node.

[0038] The term "terminal device" refers to any terminal device capable of wireless communication. As an example and not limitation, a terminal device may also be referred to as a communication device, a user equipment (UE), a subscriber station (SS), a portable subscriber station, a mobile station (MS) or an access terminal (AT). The terminal device may include, but is not limited to, a mobile phone, a cellular phone, a smart phone, a voice over IP (VoIP) phone, a wireless local loop phone, a tablet computer, a wearable terminal device, a personal digital assistant (PDA), a portable computer, a desktop computer, an image capture terminal device such as a digital camera, a game terminal device, a music storage and playback device, a vehicle-mounted wireless terminal device, a wireless endpoint, a mobile station, a laptop embedded equipment (LEE), a laptop mounted equipment (LME), a USB dongle, a smart device, a wireless customer premises equipment (CPE), an Internet of Things (IoT) device, a watch or other wearable device, a head mounted display (HMD), a vehicle, a drone, medical equipment and applications (e.g., remote surgery), industrial equipment and applications (e.g., robots and / or other wireless devices operating in the context of an industrial and / or automated processing chain), consumer electronic devices, equipment operating on a commercial and / or industrial wireless network, etc. The terminal device may also correspond to the mobile terminal (MT) part of an IAB node (eg, a relay node).In the following description, the terms "terminal device", "communication device", "terminal", "user equipment" and "UE" may be used interchangeably.

[0039] As used herein, the term "resource", "transmission resource", "resource block", "physical resource block" (PRB), "uplink resource" or "downlink resource" may refer to any resource used to perform communication (e.g., communication between a terminal device and a network device), such as a resource in the time domain, a resource in the frequency domain, a resource in the spatial domain, a resource in the code domain, or any other resource supporting communication. In the following, unless explicitly stated, resources in the frequency domain and the time domain will be used as examples of transmission resources for describing some example embodiments of the present disclosure. Note that the example embodiments of the present disclosure are also applicable to other resources in other domains.

[0040] As used herein, the term "model" is referred to as the association between an input and an output learned from training data, and therefore the corresponding output can be generated for a given input after training. The generation of the model can be based on machine learning (ML) technology. Machine learning technology can also be referred to as artificial intelligence (AI) technology. Typically, a machine learning model can be created that receives input information and makes predictions based on the input information. For example, a classification model can predict the category of input information in a predetermined number of categories. As used herein, a "model" may also be referred to as a "machine learning model," "learning model," "machine learning network," or "learning network," which may be used interchangeably herein.

[0041] Deep learning (DL) is one of the machine learning algorithms that uses multiple layers of processing units to process inputs and provide corresponding outputs. A neural network (NN) model is an example of a model based on deep learning. A neural network can process inputs to provide corresponding outputs, and typically includes an input layer, an output layer, and one or more hidden layers between the input layer and the output layer. The neural network used in deep learning typically includes a large number of hidden layers to increase the depth of the network. The layers of the neural network are connected in sequence so that the output of the previous layer is provided as the input of the next layer, wherein the input layer receives the input of the neural network, and the output of the output layer is regarded as the final output of the neural network. Each layer of the neural network includes one or more nodes (also referred to as processing nodes or neurons), each of which processes the input from the previous layer.

[0042] Typically, model lifecycle management can generally include three stages, namely the training stage, the verification stage, and the application stage (also known as the reasoning stage). In the training stage, a given machine learning model can be iteratively trained (or optimized) using a large amount of training data until the model can obtain consistent reasoning similar to that of human intelligence from the training data. During training, a set of parameter values ​​of the model is iteratively updated until the training goal is reached. Through the training process, the machine learning model can be regarded as being able to learn the association between input and output (also known as input-output mapping) from the training data. In the verification stage, the verification input is applied to the trained machine learning model to test whether the model can provide the correct output to determine the performance of the model. Typically, the verification stage can be considered as a step in the training process, or can be omitted in some cases. In the reasoning stage, the resulting machine learning model can be used to process the real-world model input and determine the corresponding model output based on a set of parameter values ​​obtained from the training process. In some cases, a retraining or update stage can be included in the model lifecycle management to enable the model to evolve to have better performance. Example Environment

[0043] Figure 1 An example communication environment 100 is shown in which example embodiments of the present disclosure may be implemented. In the communication environment 100, multiple communication devices are involved, including one or more first devices 110-1, 110-2, and 110-3, a second device 120, a third device 130, and a fourth device 140. For the purpose of discussion, the first devices 110-1, 110-2, and 110-3 are collectively or individually referred to as first devices 110.

[0044] It should be understood that Figure 1 The number of devices and their connections shown in the figure is for illustrative purposes only and does not imply any limitation. The communication environment 100 may include any suitable number of devices suitable for implementing the embodiments of the present disclosure. Although not shown, it should be understood that one or more additional devices may be involved in the communication environment 100.

[0045] Communications in the communication environment 100 may be implemented according to any suitable communication protocol, including but not limited to cellular communication protocols of the first generation (1G), second generation (2G), third generation (3G), fourth generation (4G) and fifth generation (5G), wireless local area network communication protocols such as Institute of Electrical and Electronics Engineers (IEEE) 802.11, and / or any other protocol currently known or developed in the future. In addition, communications may utilize any suitable wireless communication technology, including but not limited to: code division multiple access (CDMA), frequency division multiple access (FDMA), time division multiple access (TDMA), frequency division duplex (FDD), time division duplex (TDD), multiple input multiple output (MIMO), orthogonal frequency division multiple access (OFDM), discrete Fourier transform spread OFDM (DFT-s-OFDM) and / or any other technology currently known or developed in the future.

[0046] In the communication environment 100, the first device 110 and the fourth device 140 can communicate with each other. Figure 1 In the example of , the first device 110 is shown as a terminal device, and the fourth device 140 is shown as a network device, such as a transmission reception point (TRP). In some example embodiments, if the first device 110 is a terminal device and the fourth device 140 is a network device, the link from the fourth device 140 to the first device 110 is referred to as a downlink (DL), and the link from the first device 110 to the fourth device 140 is referred to as an uplink (UL).

[0047] Positioning techniques may be applied to obtain location information of the first device 110. In some example embodiments, the positioning techniques may be based on DL and DL plus UL positioning measurements for UE-assisted positioning performed at the first device 110 or UL and DL plus UL measurements for network-assisted positioning performed at the fourth device 140. In some cases, depending on the category of the communication channel between the first device 110 and the fourth device 140, different positioning techniques may be applied to ensure positioning accuracy. For example, to identify whether the communication channel has line-of-sight (LOS) propagation or non-line-of-sight (NLOS) propagation, different positioning methods may be applied. Therefore, it is important to classify the category of the communication channel, at least because it has an impact on the accuracy of the positioning estimate, and improving the classification accuracy can therefore improve the overall positioning accuracy.

[0048] It is proposed to deploy one or more classification models at the first device 110 to predict the classification result of the communication channel between the first device 110 and the fourth device 140. The classification model can be created based on AI technology. The processing of the classification model can be represented as Among them, f AI represents the classification model, x represents the channel measurement information related to the communication channel, Represents the estimated classification result predicted by the classification model to indicate the category into which the communication channel is classified.

[0049] First device 110 may detect a reference signal such as a positioning reference signal (PRS) propagated from fourth device 140 on a corresponding communication channel to obtain channel measurement information. In some example embodiments, fourth device 140 may transmit the PRS according to a channel classification request from second device 120. In some example embodiments, the communication channel may be classified as a LOS channel (with LOS propagation) or a NLOS channel (with NLOS propagation) through a classification model.

[0050] In some example embodiments, the second device 120 may maintain and manage the classification model used by the first device 110. The second device 120 may include a location server or controller. In some example embodiments, the second device 120 may include a network element in a core network (CN) configured for location management. In some example embodiments, the second device 120 may include a location management function (LMF), but other terms may be used.

[0051] Before the deployment of any classification model for channel classification function, there is a model training phase that usually needs to manage the following three aspects, including field channel measurement and labeling, training dataset construction and maintenance, and model online training. Online training can include direct online training and offline to online retraining (or updating).

[0052] A pre-requisite in supervised learning of models is the need for pre-labeled training data. For a classification model configured for channel classification, the labeled training data may include sample channel measurement information as sample model input and true classification results as true model labels. In some example embodiments, external gear / equipment may be deployed to support field measurements and labeling.

[0053] For example, the third device 130 in the communication environment 100 may be configured to facilitate field measurements and classification marking. The third device 130 is generally capable of determining its location. In some example embodiments, the third device 130 may include a positioning reference unit (PRU), but other terms may be used. This third device 130 may be requested by the second device 120 to perform field measurements and determine the true classification result of the channel measurement information measured at the location. It should be noted that although a third device is shown, there may be multiple third devices requested to perform classification marking. In some cases, the classification model may be updated or fine-tuned to have better performance (e.g., higher accuracy) even if the model has been deployed at the device. Such model updates may require additional labeled training data.

[0054] Training efficiency and model performance depend on training data. Informative training data contain more representative features related to the channel, and are more effective in training classification models and improving model performance relative to those with less information. However, there is no method for selecting informative training data for training classification models. Large-scale available training data can be fed indiscriminately to classification models for training purposes. Model performance increases at the cost of high computational and time resources. In some cases with high sensitivity to the cost of field data collection, reporting, and transmission, for example, by requesting the PRU to perform field classification labeling, training data may need to be carefully evaluated and selected before being sent for model updates. Otherwise, it may result in a waste of resources due to the need for the PRU to mark areas where the current classification model may have provided degraded estimation results, or miss blind spots where more training samples are needed for model training. Working principle and example process

[0055] According to some example embodiments of the present disclosure, a solution for efficient data update for channel classification is provided. In the solution, a device determines an importance level of channel measurement information related to a communication channel for updating a classification model. The importance level is compared with an importance threshold of a current classification model. If the importance level exceeds the importance threshold, a true classification result of the communication channel is obtained. The communication measurement information and the true classification result form a training data pair for updating the classification model.

[0056] Through this scheme, high-quality and low-quality data are treated differently according to their importance in updating the classification model. In this way, a small set of high-quality training data that is considered to be informative and important can be used to update the classification model. Model performance can be effectively improved with limited overhead required for classification labeling and model updating.

[0057] Hereinafter, exemplary embodiments of the present disclosure will be described in detail with reference to the accompanying drawings.

[0058] Reference now Figure 2 , which shows a flowchart of a process 200 for efficient data update of a classification model according to some example embodiments of the present disclosure. The process 200 can be implemented at any suitable device, such as Figure 1 Devices in the communication environment 100.

[0059] At block 210 , a device (eg, the first device 110 or the second device 120 ) determines an importance level of channel measurement information related to a communication channel for updating a classification model.

[0060] In some example embodiments, the communication channel to be classified may be a channel for signal propagation between first device 110 and fourth device 140. In some example embodiments, fourth device 140 may transmit a reference signal (e.g., a PRS), and first device 110 may measure the reference signal propagated on the communication channel between first device 110 and fourth device 140 to obtain channel measurement information. In some example embodiments, first device 110 may include a terminal device, and fourth device 140 may include a network device.

[0061] Channel measurement information (denoted as x i ) may include one or more types of information that helps characterize the communication channel. In some example embodiments, the channel measurement information x i The channel measurement information x may include a channel impulse response (CIR), a channel state information (CSI), a received signal strength indication (RSSI), a reference signal received power (RSRP), and / or other information that can be measured. i It can be two-dimensional information, including spatial dimension and time domain dimension. The spatial dimension can be determined by the number of antennas at the transmitter of the PRS, and the time domain dimension can be set to be greater than the maximum delay spread of wireless signal propagation. The form of channel measurement information can be configured in other ways, and the scope of the present disclosure is not limited.

[0062] The classification model is configured to determine an estimated classification result of the communication channel based at least in part on the channel measurement information. The channel classification implemented by the classification model can be expressed as: where f AI represents the classification model, x i represents channel measurement information related to the communication channel, Represents the estimated classification result.

[0063] The classification model can be constructed to extract representative features of channel measurement information in a high-dimensional feature space via machine learning, and use these features to classify communication channels. The classification model can be configured with multiple potential channel categories into which the communication channel can be classified. In some example embodiments, the classification model can perform two-class classification to classify the communication channel into a first channel category or a second channel category. In some example embodiments, the multiple channel categories may include LOS channels and NLOS channels. The estimated classification result may indicate the predicted probability that the communication channel is classified as a LOS channel or a NLOS channel. Depending on the actual application, other channel categories may also be defined, and the scope of the present disclosure is not limited.

[0064] In some cases, from the perspective of lifecycle management of the classification model, it is expected that the classification model can be updated or fine-tuned to have higher communication channel accuracy. The training data for updating the classification model can be collected on site. In the example embodiment of the present disclosure, the importance of measuring channel measurement information is proposed to determine whether the channel measurement information is used to update the classification model. As used herein, "updating" or "fine-tuning" the classification model means that a training or retraining process is triggered to update the parameter values ​​of the classification model, thereby allowing the classification model to provide more accurate classification results.

[0065] On the one hand, although the performance of the classification model can be improved if all training data are collected for updating, the training efficiency may be low if the training data cannot contribute new features and information to the classification model.

[0066] As an example, in Figure 1 , assuming that for the communication channel between the first device 110-1 and the fourth device 140, the classification model generates an estimated classification result based on the channel measurement information x1 It indicates that the communication channel is a LOS channel. For the communication channel between the first device 110-2 and the fourth device 140, the classification model generates an estimated classification result based on the channel measurement information x2. This indicates that the communication channel is an NLOS channel. For the communication channel between the first device 110-3 and the fourth device 140, the classification model generates an estimated classification result based on the channel measurement information x3. This indicates that the communication channel is a LOS channel.

[0067] By considering the actual propagation conditions of the communication channel, it can be determined that the estimated classification result at the first device 110-1 is correct and reliable, while the estimated classification results at the first devices 110-2 and 110-3 are suspicious. From the perspective of model training efficiency, it is expected that channel measurement information related to the communication channel between the first devices 110-2 and 110-3 and the fourth device is used to update the classification model to allow the model to learn and provide correct outputs for other communication channels with similar characteristics.

[0068] On the other hand, in order to form part of the training data, classification labeling is required for the communication measurement information. The cost of classification labeling is relatively high. For example, a separate device (e.g., the third device 130) may be requested to perform field labeling on the corresponding communication channel to determine the actual channel category of the communication channel as a true classification result.

[0069] Without importance analysis, it may result in wasted resources of updating the classification model with unnecessary training data, or inappropriate requests for classification labels for unnecessary data (e.g., x1) and / or missing blind spots that require classification labels (e.g., for x2 and x3).

[0070] In an example embodiment of the present disclosure, the importance level of the channel measurement information is evaluated to measure whether the channel measurement information is important when updating the classification model. Channel measurement information that is important when updating the classification model may relate to the case where the channel measurement information is informative and provides new features that are not currently captured by the classification model. In this case, the classification labels of the corresponding communication channels may provide new informative training data to fine-tune the classification model, for example, to correctly classify communication channels with similar characteristics.

[0071] There may be different types of classification models configured for channel classification tasks. For different types of classification models, the importance level of the channel measurement information may be determined in different ways. In some example embodiments, intermediate information extracted from the channel measurement information by the classification model may be directly used to determine the importance level of the channel measurement information. In some example embodiments, different intermediate information may be extracted to estimate the importance level of different types of classification models. By making full use of the intermediate information, the computational complexity for determining the importance level may be effectively reduced.

[0072] In order to measure whether the channel measurement information is important when updating the classification model, in some example embodiments, the importance level of the channel measurement information may be determined based on the uncertainty or credibility of the estimated classification result determined by the classification model based on the channel measurement information. In some example embodiments, the uncertainty level of the estimated classification result may be determined based on the intermediate information extracted by the classification model. The uncertainty level may represent the confidence or suspicion of the classification model in its estimated classification result. The uncertainty level may also be referred to as the suspicion level. Alternatively, instead of using the term "uncertainty level", the certainty level, reliability level, or confidence level of the estimated classification result may be measured.

[0073] Generally, it cannot directly identify the root cause of the classification error between "ambiguity of the channel itself" and "estimated immaturity of the classification model". However, the immaturity of the classification model has not been taken into account, which will likely mislead incorrect subsequent actions. For example, if the classification ambiguity is caused by the ambiguity of the channel itself, and the classification model is considered to trust the estimated classification result, it may affect the choice of subsequent positioning methods between geometric types (e.g., TDOA, AOA, AOD) or fingerprint type schemes. On the other hand, if the classification ambiguity is caused by the immaturity of the classification model before it is fully trained or fine-tuned, it may affect subsequent training data enhancement and model updates in terms of model lifecycle management, that is, more labeled training data needs to be collected for model fine-tuning.

[0074] In view of the above, it is beneficial to evaluate the uncertainty or credibility of the estimated classification result given by the current classification model. In some example embodiments, there may be a correspondence between potential uncertainty levels and importance levels. Specifically, if the uncertainty level is a first uncertainty level, the importance level of the channel measurement information can be determined as the first importance level. If the uncertainty level is a second uncertainty level higher than the first uncertainty level, the importance level of the channel measurement information can be determined as the second importance level, and the second importance level is higher than the first importance level. That is, the higher the uncertainty level of the estimated classification result, the higher the importance level of the corresponding channel measurement information, indicating that the communication channel or channel measurement information may be important for updating the classification model. In some examples, the uncertainty level can be determined as the importance level of the corresponding channel measurement information. In some examples, the importance level of the corresponding channel measurement information can be determined based on one or more other factors other than the uncertainty level.

[0075] In some example embodiments, the uncertainty levels of different types of classification models may be determined in different ways by taking into account the classification schemes implemented by the classification models and thus utilizing different intermediate information. Some example embodiments for calculating the uncertainty levels of different types of classification models will be described below with reference to Figures 5 to 10B Described in detail.

[0076] In some example embodiments, in addition to or as an alternative to the factor of the uncertainty level of the estimated classification result, the importance level of the channel measurement information may be determined based on other factors that may indicate that the current communication channel of the first device 110 is informative or important for improving the classification model, for example, based on communication channels in certain geographic areas that are marked as important to the classification model.

[0077] In determining the channel measurement information x i The importance level of γi In the case of, at block 220, the device (eg, the first device 110 or the second device 120) determines the importance level γ i Whether the importance threshold of the classification model is exceeded.

[0078] The importance threshold may be any predetermined threshold level.The importance threshold may be configured to control how strictly channel measurement information used to update the classification model is evaluated as "important".

[0079] If the importance level of the channel measurement information exceeds the importance threshold for the classification model, then at block 230 , the device (eg, the first device 110 or the second device 120 ) obtains a ground-truth classification result for the communication channel.

[0080] By comparing the importance level with the importance threshold, the device estimates the necessity of updating the classification model for the true classification result of the communication channel. Specifically, if it is determined that the channel measurement information is important for updating the classification model (e.g., the importance level exceeds the importance threshold), then in order to update the classification model with the channel measurement information, the true classification result is required. The true classification result (denoted as y i ) is determined as the training target label for the communication channel and is therefore considered to be the input channel measurement information x i The correct model output. In some example embodiments, the true classification result y i A communication channel may be labeled as a first channel category (eg, a LOS channel) or a second channel category (eg, a NLOS channel).

[0081] The true classification result y i and channel measurement information x i It can be used to update the classification model f AI The training data pairs {x i ,y i Using the true classification result, the parameter values ​​of the classification model can be updated so that the classification model can provide a model output that approximates the true classification result for the input classification measurement information.

[0082] In some example embodiments, the true classification result may be determined by on-site classification marking. In some example embodiments, a special device (e.g., a third device 130 (which may be a PRU) in environment 100) may be caused to perform classification marking for channel measurement information at a location associated with a communication channel of first device 110. For example, if the channel measurement information is related to a communication channel between first device 110-1 and fourth device 140, third device 130 may perform classification marking for at least the communication channel at a location associated with first device 110-1. In some example embodiments, third device 130 may be requested to perform classification marking in an area where important channel measurement information is found.

[0083] In some example embodiments, the location where the third device 130 performs classification marking may be any location in the area where the first device 110 is located. In some example embodiments, an appropriate third device 130 located near the first device 110 may be selected for classification marking. In some example embodiments, the third device 130 may be movable and may be requested to be moved to the area where the first device 110 is located. The third device 130 has the ability to determine a true classification result for a communication channel. It should be noted that although one third device 130 is shown in the environment 100, there may be multiple third devices located in different areas.

[0084] In some example embodiments, the third device 130 may be requested by the second device 120 to perform classification marking. If the second device 120 needs to obtain a true value classification result, the second device 120 may select and request an appropriate third device 130 to perform classification marking and receive the true value classification result from the third device 130. If the first device 110 needs to obtain a true value classification result, the first device 110 may send a classification marking request to the second device 120, and then the second device 120 may select and request an appropriate third device 130 to perform classification marking. In this case, the true value classification result may be forwarded by the second device 120 to the first device 110, or may be directly sent from the third device 130 to the first device 110. In some example embodiments, the first device 110 may directly request the third device 130 to assist in determining the true value classification result.

[0085] After obtaining the true classification result, at block 240, the device (eg, the first device 110 or the second device 120) causes the classification model f AI The method is updated based at least in part on the channel measurement information and the ground-truth classification result.

[0086] In some example embodiments, the update of the classification model may be performed locally on the first device 110 or the second device 120. In some example embodiments, the first device 110 as a terminal device may wish to save its computing resources, and may therefore request the second device 120 or other device to perform a model update. In this case, the first device 110 may provide the channel measurement information and the true classification result to the second device 120 or other device. Any appropriate model update or training technique may be applied to the classification model, and the scope of the present disclosure is not limited thereto.

[0087] In some example embodiments, channel measurement information and true value classification results may be added to a training data set for a classification model. A device (e.g., a first device 110 or a second device 120) may trigger an update of a classification model after collecting enough training data. For example, a device (e.g., a first device 110 or a second device 120) may determine whether the size of the newly collected training data exceeds a threshold size. If the size exceeds the threshold size, an update of the classification model may be triggered. In some example embodiments, a second device 120 or a first device 110 may collect channel measurement information and corresponding true value classification results from multiple first devices 110 to train a classification model. The collected channel measurement information may be determined to have a corresponding importance level higher than an importance threshold.

[0088] Since the training data is assessed to be important and informative, the training of the classification model can be accelerated using a small set of training data to be labeled for data efficient training, and the updated classification model can be improved to have higher accuracy.

[0089] In some cases, if the importance level of the channel measurement information is determined to be below the importance threshold, the channel measurement information x i It may not be used to update the classification model. In addition, for the channel measurement information x i The classification mark can be omitted. In this way, both the annotation efficiency and the model update efficiency are improved.

[0090] In some example embodiments, process 200 may be performed in an iterative manner. In each iteration, the classification model updated from the previous iteration may be deployed to estimate the classification results of the next iteration. For example, in the tth round, denoted as f AI(t) The classification model can be deployed and updated as denoted by f AI(t+1) The updated classification model. Classification model f AI(t+1) Can be further updated in a similar manner.

[0091] In some example embodiments, the updating of the classification model may be completed when certain criteria are met. The criteria may include criteria used for model training, criteria that no more channel measurement information is determined to have an importance level above an importance threshold (and therefore no longer requesting on-demand classification marking), or a combination of the above.

[0092] In some example embodiments, the importance threshold may be adaptively updated when the classification model is updated. AI(t) , importance threshold γ th(t) is used to compare with the classification model f AI(t) The importance levels of the classified channel measurement information are compared. If the classification model f AI(t) Updated to f AI(t+1) , then the importance threshold γ th(t) Updated to γ th(t+1) , which is then used to compare with the classification model f AI(t+1) The importance levels of the classified channel measurement information are compared.

[0093] In some example embodiments, the importance threshold may increase as the accuracy level of the classification model increases. After the updated classification model is acquired, the accuracy level of the updated classification model may be determined and used to update the importance threshold to obtain an updated importance threshold for the updated classification model. If the accuracy level of the updated classification model is higher than the accuracy level of the previously used classification model, the importance threshold may be increased by a predetermined value. The incremental value of the importance threshold may be configured as a fixed value, or may be determined based on an increase in the accuracy level. A higher increase in model accuracy may result in a higher increase in the importance threshold.

[0094] For example, if the classification model has a low accuracy level, e.g., 55%, in the initial stage, it generally means that the model may have difficulty distinguishing channel categories. The importance threshold may be set to a relatively low value so that more channel measurement information may be evaluated as important to allow more training data to be collected for model updates. As the classification model is updated and becomes more mature, the classification model accuracy level may increase, and the importance threshold may also be set to a larger value. For example, if the accuracy level of the classification model has climbed to 80%, the classification model may have more confidence in its classification results, and the importance threshold may also be increased.

[0095] The updated importance threshold can be used in the next round of model update to determine whether the channel measurement information estimated by the updated classification model is important for updating the model. By dynamically updating the importance threshold, important and informative training data for the classification model can be filtered out for efficient model update.

[0096] In some cases, in addition to the classification model used to estimate the current communication channel, there may be one or more other classification models that are deployed or to be deployed for channel classification. The collected channel measurement information and the corresponding true classification results can be shared between the classification models to update or train one or more other classification models. These classification models may have different types and / or different model configurations, but may all be configured to classify communication channels. Channel measurement information that is considered important when updating one classification model may also be important and useful when updating other classification models.

[0097] In some example embodiments, if the inputs of different classification models are different (eg, require different channel measurement information inputs), the third device 130 may be requested to collect different channel measurement information associated with the same communication channel having a true classification result.

[0098] The process of updating the classification model using the importance measure of training data has been discussed above. As mentioned above, the process can be implemented at any suitable device, which can obtain channel measurement information to measure its importance and update the classification model using the true value classification result and the channel measurement information. Some examples of this device are a first device 110 (e.g., a terminal device) measuring a communication channel, a second device 120 (e.g., a LMF) for training and managing a classification model, or even a fourth device (e.g., a network device). Some example signaling flows associated with different implementation devices will be discussed in detail below. Example signaling flow

[0099] Figure 3 3 shows a signaling flow 300 for communication according to some example embodiments of the present disclosure. Figure 3 As shown, the signaling flow 300 involves the first device 110, the second device 120, the third device 130, and the fourth device 140. For the purpose of discussion, the signaling flow 300 refers to Figure 1 Although a first device 110, a third device 130 and a fourth device 140 are Figure 3 , but it should be understood that there may be multiple first devices that perform operations similar to those described below with respect to the first device 110 and multiple third devices that perform operations similar to those described below with respect to the third device 130.

[0100] exist Figure 3 In the example embodiment of , it is assumed that the classification model for communication channel classification is deployed at the second device 120, and the second device 120 is configured for lifecycle management of the classification model. The second device 120 obtains 305 the classification model, denoted as f AI .

[0101] In some example embodiments, the second device 120 may be configured to initialize parameter values ​​of the classification model and pre-train the classification model using an available pre-stored training data set. The training data set may include a plurality of training data pairs, each pair including channel measurement information about a particular communication channel and a classification result for a communication channel label. Using the training data set, the classification model may be trained to have the ability to provide preliminary classification for the communication channel. The classification model may be immature and may require further fine-tuning to have higher accuracy.

[0102] The process 310 is implemented to collect training data for updating the classification model. Specifically, the fourth device 140 sends 315 a PRS to the first device 110. The PRS is propagated to the first device 110 via the communication channel. The first device 110 measures 320 the PRS to obtain channel measurement information (denoted as x) about the communication channel. i ).

[0103] In some example embodiments, second device 120 may request channel classification by sending a channel classification request to fourth device 140 by sending a PRS. In some example embodiments, although not specifically shown, second device 120 may request more than one fourth device 140 to send a PRS for channel classification, and fourth device 140 may send a PRS to more than one first device 110.

[0104] The first device 110 sends 325 channel measurement information x to the second device 120 i After receiving 330 channel measurement information x i When , the second device 120 then determines 335 the channel measurement information x for updating the classification model i The importance level δ i During the determination of the importance level, a classification model can be used based on the channel measurement information x i Output the estimated classification result of the communication channel Used to output estimated classification results The classification model is denoted as f AI(t) First, the classification model trained by the pre-stored training dataset is denoted as f AI(0) In some example embodiments, the classification model may then be updated iteratively according to process 310 .

[0105] In some example embodiments, instead of Figure 3Instead of directly sending the channel measurement information as shown, the first device 110 may send importance evaluation information, such as the importance level or intermediate information for determining the importance level, to the second device 120. In this case, the classification model may be deployed at the first device 110.

[0106] In some example embodiments, for some types of classification models that require low overhead for calculating the importance level, first device 110 may determine the importance level of the channel measurement information and send the importance level to second device 120. In some example embodiments, for some other types of classification models that require high overhead, first device 110 may request other devices (such as second device 120) to assist in determining the importance level. In this case, first device 110 may provide channel measurement information or intermediate information to second device 120 to determine the importance level.

[0107] Using the determined importance level δ i , the second device 120 determines the channel measurement information x i The importance level δ i Whether it exceeds the classification model f AI(t) The importance threshold δ th(t) If the importance level δ i Exceeding the importance threshold δ th(t) (For example, δ i >δ th(t) ), the second device 120 determines to obtain a true classification result for the communication channel. Specifically, the second device 120 sends 340 a classification tag request to an appropriate third device 130 at a location associated with the first device 110. By receiving 345 the classification tag request, the third device 130 classifies and tags the communication channel between the first device 110 and the fourth device 140 to determine the correct channel category of the communication channel. The fourth device 140 sends 350 a true classification result y to the second device 120. i ,y i Indicates the correct channel category. True classification result y i Can be used as channel measurement information x i The target training labels.

[0108] After receiving 355 the true classification result, the second device 120 uses at least the channel measurement information and the true classification result {x i ,y i} Classification model f AI(t) Update 360 ​​to classification model f AI(t+1) In some example embodiments, the second device 120 may trigger the classification model f after multiple pairs of channel measurement information and corresponding true classification results are obtained. AI(t) Updates.

[0109] In some example embodiments, the second device 120 will target the classification model f AI(t) The importance threshold δ th(t) Update 365 is for the updated classification model f AI(t+1) The updated importance threshold δ th(t+1) The updated importance threshold can be based on the updated classification model f AI(t+1) The accuracy level is determined, for example, based on the classification model f AI(t) To the updated classification model f AI(t+1) The accuracy level increases.

[0110] In some cases, if the channel measurement information x i The importance level δ i is determined to be below the importance threshold, the second device 120 may discard the channel measurement information x i In this way, classification marking is not triggered for channel measurement information that is not important when updating the model. Both marking efficiency and model updating efficiency are improved.

[0111] In some example embodiments, process 310 may be iteratively performed to fine-tune the classification model until the model update is complete. The update of the classification model may be completed when a certain criterion is met. The criterion may include a criterion for model training, a criterion that no more channel measurement information is determined to have an importance level above an importance threshold (and therefore no longer requesting on-demand classification marking), or a combination of the above.

[0112] In some example embodiments, it is assumed that a classification model for communication channel classification is deployed at the first device 110, and the first device 110 is configured for lifecycle management of the classification model. Figure 4 A signaling flow 400 according to such an example embodiment is shown.

[0113] like Figure 4 As shown, the signaling flow 400 involves the first device 110, the second device 120, the third device 130, and the fourth device 140. For the purpose of discussion, the signaling flow 400 refers to Figure 1 Although a first device 110, a third device 130 and a fourth device 140 are Figure 4 , but it should be understood that there may be multiple first devices that perform operations similar to those described below with respect to the first device 110 and multiple third devices that perform operations similar to those described below with respect to the third device 130.

[0114] exist Figure 4In the example embodiment of , the second device 120 provides 405 the classification model to the first device 110, denoted as f AI The first device 110 downloads 410 the classification model from the second device 120 .

[0115] In some example embodiments, second device 120 may provide a classification model upon receiving a positioning service request from first device 110. The classification model may have been trained by second device 120 using a pre-stored training data set or a training data set collected partially or entirely from field data, for example, via signaling flow 300.

[0116] Process 412 is implemented to collect training data for updating a classification model deployed at first device 110. Specifically, fourth device 140 sends 415 a PRS to first device 110. The PRS is propagated to first device 110 via a communication channel. In some example embodiments, second device 120 may request channel classification by sending a channel classification request to fourth device 140 by sending the PRS.

[0117] The first device 110 measures 420 the PRS to obtain channel measurement information (denoted as x) about the communication channel. i ). The first device 110 may use the classification model to determine the channel measurement information x based on the channel measurement information x. i Determine the estimated classification result of the communication channel Used to output estimated classification results The classification model is denoted as f AI(t) First, the classification model trained by the pre-stored training dataset is denoted as f AI(0) In some example embodiments, the classification model may then be iteratively updated according to process 412 .

[0118] The first device 110 determines 425 channel measurement information x for updating the classification model i The importance level δ i In some example embodiments, considering the workload of determining the importance level of a specific type of classification model as described above, the first device 110 may request other devices (such as the second device 120) to determine the importance level by providing channel measurement information and / or other necessary information to the other devices.

[0119] The first device 110 sets the importance level δ i With the classification model f AI(t) The importance threshold δ th(t) If the importance level δ i Exceeding the importance threshold δ th(t) (For example, δ i >δ th(t)), the first device 110 determines to obtain a true classification result for the communication channel. In this case, the first device 110 triggers 430 a sub-process for performing classification marking on the communication channel.

[0120] In some example embodiments, in the sub-process of classification marking, the first device 110 may send a classification marking request to the second device 120. Then, the second device 120 may select and request an appropriate third device 130 to perform classification marking at a location associated with the first device 110. In some example embodiments, in the sub-process of classification marking, the first device 110 may directly request the third device 130 to perform classification marking.

[0121] The first device 110 receives 440 the true classification result y sent 435 from the third device 130 i ,y i Indicates the channel category of the communication channel between the first device 110 and the fourth device 140. True classification result y i Can be used as channel measurement information x i In some example embodiments, the third device 130 may classify the true value result y i The first device 110 may receive the true classification result y forwarded by the second device 120. i .

[0122] The first device 110 uses at least the channel measurement information and the true value classification result {x i ,y i} Classification model f AI(t) Update 445 for classification model f AI(t+1) In some example embodiments, the first device 110 may trigger the classification model f after multiple pairs of channel measurement information and corresponding true classification results are obtained. AI(t) Updates.

[0123] In some example embodiments, the first device 110 will target the classification model f AI(t) The importance threshold δ th(t) Update 450 is for the updated classification model f AI(t+1) The updated importance threshold δ th(t+1) The updated importance threshold can be based on the updated classification model f AI(t+1) The accuracy level is determined, for example, based on the classification model f AI(t) To the updated classification model f AI(t+1) The accuracy level increases.

[0124] In some example embodiments, instead of locally updating the classification model and / or the importance threshold, the first device 110 may cause the update to be performed by other devices, such as the second device 120. The first device 110 may provide channel measurement information and / or true value classification results to facilitate updating the classification model and then updating the importance threshold. The scope of the present disclosure is not limited in this regard.

[0125] In some cases, if the channel measurement information x i The importance level δ i is determined to be below the importance threshold, the first device 110 may discard the channel measurement information x i In this way, classification marking is not triggered for channel measurement information that is not important when updating the model. Both marking efficiency and model updating efficiency are improved.

[0126] In some example embodiments, process 412 may be iteratively performed to fine-tune the classification model until the model update is complete. The update of the classification model may be completed when a certain criterion is met. The criterion may include a criterion for model training, a criterion that no more channel measurement information is determined to have an importance level above an importance threshold (and therefore no longer requesting on-demand classification marking), or a combination of the above.

[0127] The update of the classification model for channel classification has been described above from the perspective of second device 120 or first device 110. In some example embodiments, fourth device 140 may acquire channel measurement information, for example, by receiving the channel measurement information from first device 110 or by measuring a reference signal sent from first device 110. In this case, if the classification model is deployed at fourth device 140, fourth device 140 may perform similar operations as described herein with respect to first device 110 to update the classification model. Example of classification model and uncertainty level calculation

[0128] As described above, based on the type of classification model, the uncertainty level of the estimated classification result (and thus the importance level of the channel measurement information) may be determined.

[0129] In some example embodiments, for a particular type of classification model, an uncertainty level of an estimated classification result may be determined by reconstructing the classification model. An example of such a classification model is a deep neural network (DNN), which typically provides a hard output (e.g., 0 or 1) to indicate whether a communication channel is classified as a first channel category or a second channel category. For such a type of classification model, multiple reference classification models may be generated by reconstructing the classification model and used to determine the uncertainty level.

[0130] Figure 5A flowchart of a process 500 for determining an uncertainty level of an estimated classification result output by a DNN type classification model according to some example embodiments of the present disclosure is shown. The process 500 can be implemented, for example, by a device that needs to determine an uncertainty level (or importance level). Such a device can include the first device 110 or the second device 120.

[0131] At block 510 , a device (eg, the first device 110 or the second device 120 ) generates a plurality of reference classification models by reconstructing a classification model.

[0132] Fig. 6A An example of a DNN type classification model 610 and a reference classification model generated therefrom according to some example embodiments of the present disclosure is shown. The classification model 610 may be in the form of a DNN model.

[0133] like Fig. 6A As shown, the classification model 610 is configured as an input layer 602, one or more intermediate layers 604, 606, and an output layer 608, each layer including a plurality of operating units (sometimes referred to as neurons). An operating unit in one layer is connected to an operating unit in the next layer. In some example embodiments, an operating unit in one layer may be connected to one or more other operating units in the same layer. Channel measurement information x i The information is input to the input layer 602 for processing, and is propagated through the interior of the intermediate layers 604 and 606 according to the connection of the layers. i The estimated classification result y i is output from the output layer 608. Examples of layers included in the classification model may include a convolution layer, a batch normalization, an activation function, a pooling layer, a fully connected layer, an LSTM (Long Short Term Memory) layer, and other types of layers.

[0134] Note that the number and number of layers of the operation units in the classification model 610 are independent of the exemplary embodiments of the present disclosure, and these numbers are given values. The structure of the classification model is also non-limiting, and the connections between the operation units can be recursive or bidirectional. Any model suitable for channel classification can be used.

[0135] For DNN type models and similar models, there is no general solution available to directly measure the uncertainty level of the estimated results based on the model output and input. It is worth noting that even if the DNN type classification model outputs a probability vector as an estimated classification result, it may not be directly used to indicate the uncertainty level of the estimated classification result. That is, even with a high output probability, the classification model may be uncertain in its prediction. In some example embodiments, it is proposed to reconstruct the classification model by generating multiple reference classification models by slightly changing the model, and determine the uncertainty level based on the multiple reference classification models.

[0136] exist Fig. 6A In the example, J reference classification models 612-1, 612-2, ..., 612-J (collectively or individually referred to as reference classification model 612) for the classification model, where J is an integer greater than 1. In some example embodiments, the classification model can be reconstructed by applying random neural connection discarding. Specifically, some neural connections between operation units in different layers of the classification model can be discarded to obtain a reference classification model. In some example embodiments, a discard ratio can be applied to determine the ratio of discarded neural connections. In some example embodiments, the second device 120 can apply a Gaussian process to determine which neural connections are discarded from the classification model. In some other example embodiments, other discarding means can also be applied to generate a reference classification model.

[0137] In some example embodiments, a reference classification model may be constructed such that the channel measurement information x i As its model input and output reference estimated classification results. In such an example embodiment, the reference classification model can have a similar structure to the classification model, except that some neural connections are discarded. The reference estimated classification result output by the jth reference classification model can be expressed as Where j=1,2,…,J.

[0138] In some example embodiments, considering that the DNN type classification model has a hierarchical structure, it can be composed of different consecutive model parts, wherein the output of the first model part can be provided for further processing in the next second model part. The intermediate information propagated between different model parts (or layers) can be referred to as feature information or latent information of the input channel measurement information. In some example embodiments, the DNN type classification model can be considered to include a feature extraction model part and a detection model part. The feature extraction model part can be implemented by artificially designed feature engineering or the machine itself via deep learning to extract feature information from the channel measurement information. The detection model part can be implemented as a neural network having one or more fully connected neurons and an output layer (such as Figure 5The detection model part can receive the feature information extracted by the feature extraction model part, and process the feature information to output an estimated classification result.

[0139] In some example embodiments, in order to reduce computational complexity, multiple reference classification models may be generated by reconstructing a post-model portion of the classification model. For example, if the classification model is divided into a first model portion (e.g., a feature extraction model portion) followed by a second model portion (e.g., a detection model portion), multiple reference classification models may be generated by reconstructing the second model portion. In some example embodiments, the second model portion to be reconstructed may include the last layer or layers of the classification model. In this way, the intermediate information output by the first model portion of the classification model (e.g., feature information extracted by the feature extraction model portion) may be reused by multiple reference classification models.

[0140] Figure 6B An example of a reference classification model generated from a second model portion of a classification model is shown. As shown, the second model portion including layers 606 and 608 of the classification model 610 can be reconstructed to generate J reference classification models 612, for example, by discarding one or more neural connections between layers 606 and 608. For example, neurons in layer 606 can be discarded in different ways to generate different layers 616-1, 616-2, ..., 616-J for the J reference classification models 612, respectively.

[0141] Return to reference Figure 5 At block 520, the device (eg, the first device 110 or the second device 120) uses a plurality of reference classification models to determine a plurality of reference estimated classification results.

[0142] In with Fig. 6A In some related example embodiments, the channel measurement information x i The channel measurement information x may be directly provided as input to the multiple reference classification models 612. The multiple reference classification models 612 may independently process the channel measurement information x i To generate the corresponding reference estimation classification results In with Figure 6B In some related example embodiments, the intermediate information extracted by the first model portion of the classification model 610 can be provided as input to multiple reference classification models 612. In this way, since the intermediate information can be reused, the computational complexity and resource cost can be reduced. The reference classification model 612 can further process the intermediate information to generate corresponding reference estimated classification results.

[0143] At block 530, a device (eg, the first device 110 or the second device 120) estimates a classification result based on a plurality of references. The variance of the channel measurement information x is determined i The estimated classification results The uncertainty level (expressed as γ i ).

[0144] In some example embodiments, if multiple reference estimation classification results The variance of is relatively high, which means that the reference classification model is inconsistent in classifying the channel measurement information xu, so the original classification model may not trust the estimated classification result In this case, the estimated classification result The uncertainty level of i can be determined to be a relatively high level, and thus the channel measurement information can be determined to be informative and important for updating the original classification model. i is considered to be of high importance, and the classification model needs to be updated to become stable and have greater confidence in classifying similar communication channels even if the model structure changes slightly (e.g., by dropping some connections).

[0145] In some example embodiments, if the variance of the multiple reference estimated classification results is relatively low, the estimated classification result The uncertainty level of i may be determined to be a relatively low level, and thus the channel measurement information may be determined to be unimportant for updating the original classification model.

[0146] The variance of the multiple reference estimated classification results may be measured in any suitable manner. In some example embodiments, the variance of the multiple reference estimated classification results may be determined based on each of the reference estimated classification results and the expectation of the reference estimated classification results. In some example embodiments, the estimated classification results can be represented as a vector, including elements indicating the respective probabilities that the communication channel is classified into the first channel category and the second channel category. In this way, the uncertainty level γ i can be determined as follows: where I is the identity matrix, and in represents the expectation, N is the number of input channel measurement information samples input to the classification model, d is the discard ratio, λ is the weight decay of the L2 adjustment term in the classification model, and l is the prior length scale describing the prior knowledge of the parameter value of the classification model.

[0147] Note that the above equation (1) is only used to calculate the uncertainty level γ based on the variance i There may be many other ways to calculate the variance and then the uncertainty level, and the scope of the present disclosure is not limited in this regard.

[0148] As described above, where the uncertainty level of the estimated classification result is determined, the importance level of the corresponding channel measurement information can be determined accordingly. As described above, the importance level can be used to select important channel measurement information for updating the classification model to allow efficient training of the model.

[0149] Figure 7 The model performance gain of some example embodiments of the present disclosure relative to the traditional model training method is shown. According to the traditional model training method, the classification labeling for the channel measurement information can be randomly performed without importance evaluation, so the training data (channel measurement information and true value classification results) are randomly used to update the DNN type classification model. According to the example embodiments of the present disclosure, important and informative channel measurement information is found to update the DNN type classification model.

[0150] Figure 7 The accuracy trend curve 710 of the proposed method according to some example embodiments of the present disclosure and the accuracy trend curve 720 of the traditional model training method are shown. Both trend curves show the accuracy climb and the number of labeled training data for training DNN type classification models. It can be seen that in order to achieve the same level of accuracy, the number of labeled training data pairs required in the example embodiments of the present disclosure can be substantially reduced compared to the traditional method. Another example of classification model and uncertainty level calculation

[0151] In some example embodiments, for a particular type of classification model, the classification result is estimated The uncertainty level of i It can be evaluated without rebuilding the classification model. In some example embodiments, some types of classification models can be provided for further deriving estimated classification results. As an example, in the case of binary classification, the classification model can be based on the channel measurement information x i To determine a first number of model votes for a first channel category (denoted as "N1") and a second number of model votes for a second channel category (denoted as "N2"). An estimated classification result can be determined based on a ratio of the first number to the second number (e.g., N1 / N2), where a higher ratio can indicate a higher probability that the communication channel is classified into the first channel category.

[0152] Some examples of this type of classification model may include, but are not limited to, a k-nearest neighbor (KNN) model, a support vector machine (SVM) model, and a random forest (RF) model. In some example embodiments, the uncertainty level of the estimated classification result output by this type of classification model may be determined based on intermediate information (e.g., a first number N1 of model votes for a first channel category and a second number N2 of model votes for a second channel category).

[0153] Figure 8 A flowchart of a process 800 for determining an uncertainty level of an estimated classification result output by a DNN type classification model according to some example embodiments of the present disclosure is shown. Process 800 can be implemented, for example, by a device that needs to determine an uncertainty level (or importance level). Such a device can include the first device 110 or the second device 120.

[0154] At block 810, a device (eg, the first device 110 or the second device 120) counts a first number N1 of model votes for a first channel category and a second number N2 of model votes for a second channel category. These two numbers can be obtained from the classification model without rebuilding the classification model.

[0155] Fig. 9A An example of a KNN type classification model according to some example embodiments of the present disclosure is shown. Fig. 9A , the feature space 900 includes a plurality of features 902 associated with a first channel category (denoted as “category 1”) and a plurality of features 904 associated with a second channel category (denoted as “category 2”). According to a KNN type classification model, the classification scheme is configured to measure the respective distances between features extracted from the channel measurement information and features in the feature space 900, and select a predetermined number (e.g., K) of features having shorter distances (e.g., K features having the shortest distances).

[0156] Among the total K selected features, the classification model can count a first number of features associated with the first channel category (i.e., a first number N1 of model votes for the first channel category) and a second number of features associated with the second channel category (i.e., a second number N2 of model votes for the second channel category). The first number and the second number can be used to determine the probability that the communication channel belongs to the first channel category.

[0157] exist Fig. 9A In the example of FIG. 1 , the channel measurement information x obtained by the first device 110 - 1 iThe feature 912 of is close to six features associated with category 1 and one feature associated with category 2. Therefore, the first probability that the communication channel is of the first channel category is 1 / 7, and the second probability that the communication channel is of the second channel category is 6 / 7. Since the second probability is higher than the first probability, the estimated classification result can be determined to indicate that the communication channel is classified into the second channel category. Since 6 / 7 is close to 100%, this indicates that the estimated uncertainty level of the result is low, x i So it is not important for updating the classification model.

[0158] In another example, the channel measurement information x j The feature 914 of is close to four features associated with category 1 and three features associated with category 2. Therefore, the first probability that the communication channel is of the first channel category is 4 / 7, and the second probability that the communication channel is of the second channel category is 3 / 7. Since the first probability (i.e., 4 / 7) is higher than the second probability, the estimated classification result can be determined to indicate that the communication channel is classified into the first channel category. However, since 4 / 7 is close to 50%, this indicates that the estimated uncertainty level of the result is high, x i It is therefore important for updating classification models.

[0159] It should be understood that Fig. 9A The example of is provided for the purpose of illustration, and does not suggest a classification method for the first type of KNN type classification model. Some classification models may operate in other ways to determine the model votes for the two channel categories and then output an estimated classification result.

[0160] Fig. 9B An example of a classification model of RF type according to some example embodiments of the present disclosure is shown. Fig. 9B , the RF type classification model 922 includes a plurality of decision trees 924-1, 924-2, 924-3, ..., 924-K (collectively or individually referred to as decision trees 924). The random feature selection part 920 is configured to select a random feature from the channel measurement information x i K subsets of information or features are randomly extracted from i1 , x i2 , x i3 , …, x iK .

[0161] Information or feature x i1 , x i2 , x i3 , …, x iKThe L subsets of can be provided as inputs to decision trees 924-1, 924-2, 924-3, ..., 924-K, respectively. Decision tree 924 can be configured to process a subset of information or features and output a decision about channel classification. Specifically, decision tree 924 can be configured with two potential outputs, one output indicating a first channel category (denoted as "category 1") and another output indicating a second channel category (denoted as "category 2"). Fig. 9B In the illustrated example of , decision trees 924 - 1 , 924 - 3 , 924 -K and possibly other decision trees may output a decision 932 regarding category 1 , and decision tree 924 - 2 and possibly other decision trees may output a decision 934 regarding category 2 .

[0162] In the RF type classification model 922, the number of decisions for category 1 is represented as N1, and the number of decisions for category 1 is represented as N2, where N1+N2=K. The estimated classification result of the communication channel can be determined by comparing N1 with N2. In some examples, according to the example strategy configured in the RF type classification model 922, if N1 is greater than N2, the estimated classification result can be determined to indicate that the communication channel is classified into the first channel category. Otherwise, if N1 is less than or equal to N2, the estimated classification result can be determined to indicate that the communication channel is classified into the second channel category.

[0163] It should be understood that the SVM type classification model can operate in a similar manner to the KNN type classification model, which will not be specifically shown. The first number of first channel categories and the second number of first channel categories can be obtained from the classification model and can therefore be used to help determine the uncertainty level of the estimated classification result.

[0164] At block 820, the device (e.g., the first device 110 or the second device 120) determines a degree of difference between the first number N1 of the first channel category and the second number N2 of the second channel category. At block 830, the device (e.g., the first device 110 or the second device 120) determines an estimated classification result based on the degree of difference. The uncertainty level of i .

[0165] In the case of model voting for binary classification, if the classification model is more confident in its estimate, the number of model votes for one channel class can be larger, and correspondingly, the number of model votes for another channel class can be smaller. Therefore, if the degree of difference between the first number and the second number is relatively high, it means that the classification model is relatively confident in its classification result, and therefore the uncertainty level of the estimated classification result may be low.

[0166] In some example embodiments, the sum of N1 and N2 is determined to be K, and the degree of difference between the first number N1 and the second number N2 may be measured based on the larger value between the ratio of N1 to K and the ratio of N2 to K, which may be expressed as min(N1, N2) / K. In some example embodiments, the degree of difference between the first number N1 and the second number N2 may be measured based on the larger value between N1 / N2 or N2 / N1, which may be expressed as min(N1 / N2, N2 / N1). In these cases, if min(N1, N2) / K or min(N1 / N2, N2 / N1) is determined to have a higher value, the uncertainty level γ i In some examples, the uncertainty level γ i can be determined as follows: γ i =min(N1, N2) / K or min(N1 / N2, N2 / N1). In other examples, the uncertainty level γ i It may be determined in other ways based on the degree of difference between N1 and N2.

[0167] There may be many other methods to calculate the uncertainty level based on the first number of first channel categories and the second number of second channel categories, and the scope of the present disclosure is not limited.

[0168] As described above, where the uncertainty level of the estimated classification result is determined, the importance level of the corresponding channel measurement information can be determined accordingly. As described above, the importance level can be used to select important channel measurement information for updating the classification model to allow efficient training of the model.

[0169] Fig. 10A The model performance gain of some example embodiments of the present disclosure for a KNN-type classification model relative to a traditional model training method is shown. Fig. 10B The model performance gain of some example embodiments of the present disclosure for the RF type classification model relative to the traditional model training method is shown. According to the traditional model training method, the classification labeling for the channel measurement information can be randomly performed without importance evaluation, so the training data (channel measurement information and true value classification results) are randomly used to update the KNN type classification model or the RF type classification model. According to the example embodiments of the present disclosure, important and informative channel measurement information is found to update the DNN type classification model or the RF type classification model.

[0170] Fig. 10AAn accuracy trend curve 1010 of the proposed method according to some example embodiments of the present disclosure and an accuracy trend curve 1020 of the traditional model training method are shown. Both trend curves show the accuracy climb and the number of labeled training data for training a KNN type classification model. It can be seen that, compared with the traditional method, using the proposed method according to some example embodiments of the present disclosure, in order to achieve the same model performance in terms of classification accuracy, the number of labeled training data required for the classification model can be reduced by approximately 60%.

[0171] Fig. 10B Also shown are the accuracy trend curve 1012 of the proposed method according to some example embodiments of the present disclosure and the accuracy trend curve 1022 of the traditional model training method. Both trend curves show the accuracy climb and the number of labeled training data for training the RF type classification model. It can be seen that, compared with the traditional method, using the proposed method according to some example embodiments of the present disclosure, in order to achieve the same model performance in terms of classification accuracy, the number of labeled training data required for the classification model can be reduced by approximately 60%. Example devices, equipment, and media

[0172] In some example embodiments, a device capable of performing any of the methods 800 (e.g., Figure 1 The first device 110 or the second device 120 in the method 800 may include a component for performing the corresponding operation of the method 800. The component may be implemented in any suitable form. For example, the component may be implemented in a circuit system or a software module. The device may be implemented as or included in Figure 1 In the first device 110 or the second device 120.

[0173] In some example embodiments, the apparatus includes: a component for determining an importance level of channel measurement information associated with a communication channel for updating a classification model, the classification model being configured to determine an estimated classification result of the communication channel based on the channel measurement information; a component for obtaining a true classification result for the communication channel if it is determined that the importance level of the channel measurement information exceeds an importance threshold of the classification model; and a component for causing the classification model to be updated at least in part based on the channel measurement information and the true classification result.

[0174] In some example embodiments, the apparatus further comprises: means for determining an accuracy level of the updated classification model; and means for updating the importance threshold based on the accuracy level to obtain an updated importance threshold for the updated classification model.

[0175] In some example embodiments, the means for updating the importance threshold based on the accuracy level includes means for increasing the importance threshold by a predetermined value if it is determined that the accuracy level of the updated classification model is higher than the accuracy level of the classification model.

[0176] In some example embodiments, the means for determining the importance level includes: means for obtaining intermediate information extracted from the channel measurement information by the classification model; and means for determining the importance level of the channel measurement information based on the intermediate information.

[0177] In some example embodiments, the component for determining the importance level based on the intermediate information includes: a component for determining the uncertainty level of the estimated classification result based on the intermediate information; a component for determining the importance level of the channel measurement information to be the first importance level if the determined uncertainty level is the first uncertainty level; and a component for determining the importance level of the channel measurement information to be the second importance level if the determined uncertainty level is a second uncertainty level higher than the first uncertainty level, the second importance level being higher than the first importance level.

[0178] In some example embodiments, a component for determining an uncertainty level of an estimated classification result includes: a component for generating multiple reference classification models by applying random neural connection discarding to the classification model if the classification model is determined to be a deep neural network; a component for determining multiple reference estimated classification results based on intermediate information using the multiple reference classification models; and a component for determining the uncertainty level based on the variance of the multiple reference classification results.

[0179] In some example embodiments, the classification model includes a first model portion and a second model portion connected to the first model portion, and the component for generating multiple reference classification models includes: a component for generating multiple reference classification models by applying random neural connection dropout to the second model portion in the classification model.

[0180] In some example embodiments, the intermediate information includes feature information extracted from the channel measurement information by a first model portion of the classification model. The means for determining the plurality of reference estimated classification results includes: means for applying the feature information as an input to the plurality of reference classification models, and means for obtaining the plurality of reference estimated classification results output from the plurality of reference classification models.

[0181] In some example embodiments, the intermediate information includes: a first number of votes cast by the classification model for a model of a first channel category based on the channel measurement information and a second number of votes cast for a model of a second channel category, and an estimated classification result is determined based on the first number and the second number. The means for determining the uncertainty level of the estimated classification result includes: means for determining a degree of difference between the first number of the first channel category and the second number of the second channel category; and means for determining the uncertainty level based on the degree of difference.

[0182] In some example embodiments, the component for obtaining the true value classification result includes: a component for causing a positioning reference device to perform classification marking on channel measurement information at a location associated with a communication channel; and a component for receiving the true value classification result from the positioning reference device.

[0183] In some example embodiments, the estimated classification result indicates whether the communication channel is classified as a line-of-sight channel or a non-line-of-sight channel.

[0184] In some example embodiments, the apparatus comprises a terminal device or a location management function.

[0185] In some example embodiments, the apparatus further comprises means for performing process 200 or other operations in some example embodiments of first or second device 110 or 120. In some example embodiments, the means comprises: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause execution of the apparatus.

[0186] Fig.11 1 is a simplified block diagram of a device 1100 suitable for implementing an example embodiment of the present disclosure. The device 1100 may be provided to implement a communication device, such as Figure 1 The first device 110, the second device 120 or other devices shown in FIG. As shown in the figure, the device 1100 includes one or more processors 1110, one or more memories 1120 coupled to the processor 1110, and one or more communication modules 1140 coupled to the processor 1110.

[0187] The communication module 1140 is used for two-way communication. The communication module 1140 has one or more communication interfaces to support communication with one or more other modules or devices. The communication interface may represent any interface necessary for communicating with other network elements. In some example embodiments, the communication module 1140 may include at least one antenna.

[0188] As non-limiting examples, processor 1110 may be of any type suitable for a local technology network, and may include one or more of the following: a general purpose computer, a special purpose computer, a microprocessor, a digital signal processor (DSP), and a processor based on a multi-core processor architecture. Device 1100 may have multiple processors, such as application specific integrated circuit chips, which are time-slaved to the clock of a synchronized master processor.

[0189] The memory 1120 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) 1124, electrically programmable read-only memory (EPROM), flash memory, hard disk, compact disk (CD), digital video disk (DVD), optical disk, laser disk, and other magnetic storage and / or optical storage. Examples of volatile memories include, but are not limited to, random access memory (RAM) 1122 and other volatile memories that do not persist during power outages.

[0190] Computer program 1130 includes computer executable instructions executed by associated processor 1110. Program 1130 may be stored in a memory, such as ROM 1124. Processor 1110 may perform any suitable actions and processes by loading program 1130 into RAM 1122.

[0191] The exemplary embodiments of the present disclosure may be implemented with the aid of the program 1130 so that the device 1100 may execute the procedures described in the reference Figures 2 to 10B Any process of the present disclosure discussed. The exemplary embodiments of the present disclosure may also be implemented by hardware or by a combination of software and hardware.

[0192] In some example embodiments, the program 1130 may be tangibly embodied in a computer-readable medium, which may be included in the device 1100 (such as in the memory 1120) or in other storage devices accessible by the device 1100. The device 1100 may load the program 1130 from the computer-readable medium to the RAM 1122 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. Fig.12 An example of a computer readable medium 1200 is shown which may be in the form of a CD, DVD, or other optical storage disk. The computer readable medium 1200 has a program 1130 stored thereon.

[0193] Generally, various embodiments of the present disclosure may be implemented in hardware or dedicated circuits, software, logic circuits, or any combination thereof. Some aspects may be implemented in hardware, while other aspects may be implemented in firmware or software that may be executed by a controller, microprocessor, or other computing device. Although various aspects of the embodiments of the present disclosure are shown and described as block diagrams, flow charts, or using some other graphical representations, it should be understood that, as non-limiting examples, the blocks, devices, systems, techniques, or methods described herein may be implemented in hardware, software, firmware, dedicated circuits or logic circuits, general-purpose hardware or controllers or other computing devices, or some combination thereof.

[0194] The present disclosure also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. As used herein, the term "non-transitory" is a limitation of the medium itself (i.e., tangible, not a signal), not a limitation on the persistence of data storage (e.g., RAM vs. ROM). The computer program product includes computer executable instructions, such as included in a program module, executed in a device on a target physical or virtual processor to perform the above referenced Figures 2 to 10B Those instructions of any method described.Usually, program modules include routines, programs, libraries, objects, classes, components, data structures, etc. that perform specific tasks or implement specific abstract data types.In various embodiments, the functions of program modules can be combined or split between program modules as needed.Machine executable instructions for program modules can be executed in local or distributed devices.In distributed devices, program modules can be located in both local and remote storage media.

[0195] The program code for executing the method of the present disclosure can be written in any combination of one or more programming languages. The program code can be provided to a processor or controller of a general-purpose computer, a special-purpose computer or other programmable data processing device so that the program code enables the functions / operations specified in the flow chart and / or block diagram to be implemented when executed by the processor or controller. The program code can be executed entirely on the machine, partially on the machine, as an independent software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0196] In the context of the present disclosure, computer program codes or related data may be carried by any suitable carrier to enable a device, apparatus or processor to perform various processes and operations as described above. Examples of carriers include signals, computer readable media, etc.

[0197] The computer readable medium may be a computer readable signal medium or a computer readable storage medium. The computer readable medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared or semiconductor systems, devices or apparatuses, or any suitable combination of the foregoing. More specific examples of computer readable storage media would include an electrical connection with one or more wires, a portable computer disk, 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 disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0198] In addition, although the operation is depicted in a particular order, this should not be understood as requiring such operation to be performed in the particular order shown or in order, or performing all the operations shown to achieve the desired result. In some cases, multitasking and parallel processing may be advantageous. Similarly, although several specific implementation details are included in the above discussion, these details should not be interpreted as limiting the scope of the present disclosure, but should be interpreted as a specific description of the features that can be to a particular embodiment. The specific features described in the context of a single embodiment can also be implemented in combination in a single embodiment. On the contrary, the various features described in the context of a single embodiment can also be implemented in multiple embodiments individually or in any suitable sub-combination.

[0199] Although the disclosure has been described in language specific to structural features and / or methodological acts, it should be understood that the 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 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 device to at least perform: determining an importance level of channel measurement information associated with a communication channel for updating a classification model, the classification model being configured to determine an estimated classification result of the communication channel based on the channel measurement information; If it is determined that the importance level of the channel measurement information exceeds the importance threshold for the classification model, obtaining a true value classification result for the communication channel; as well as The classification model is caused to be updated based at least in part on the channel measurement information and the ground-truth classification result.

2. The apparatus according to claim 1, wherein the apparatus is further caused to execute: determining a level of accuracy of the updated classification model; and The importance threshold is updated based on the accuracy level to obtain an updated importance threshold for the updated classification model.

3. The apparatus of claim 2, wherein updating the importance threshold based on the accuracy level comprises: If it is determined that the accuracy level of the updated classification model is higher than the accuracy level of the classification model, the importance threshold is increased by a predetermined value.

4. The apparatus according to any one of claims 1 to 3, wherein determining the importance level comprises: acquiring intermediate information extracted by the classification model from the channel measurement information; as well as The importance level of the channel measurement information is determined based on the intermediate information.

5. The apparatus of claim 4, wherein determining the importance level based on the intermediate information comprises: determining an uncertainty level of the estimated classification result based on the intermediate information; If it is determined that the uncertainty level is the first uncertainty level, determining the importance level of the channel measurement information to be the first importance level; as well as If it is determined that the uncertainty level is a second uncertainty level higher than the first uncertainty level, the importance level of the channel measurement information is determined to be a second importance level, and the second importance level is higher than the first importance level.

6. The apparatus of claim 5, wherein determining the level of uncertainty of the estimated classification result comprises: If it is determined that the classification model is a deep neural network, generating a plurality of reference classification models by applying random neural connection dropout to the classification model; using the plurality of reference classification models, determining a plurality of reference estimated classification results based on the intermediate information; as well as The uncertainty level is determined based on a variance of the plurality of reference classification results.

7. The apparatus of claim 6, wherein the classification model comprises a first model portion and a second model portion connected to the first model portion, and wherein generating the plurality of reference classification models comprises: The plurality of reference classification models are generated by applying random neural connection dropout to the second model portion in the classification model.

8. The apparatus of claim 7, wherein the intermediate information comprises feature information extracted from the channel measurement information by the first model portion in the classification model; and Wherein determining the plurality of reference estimation classification results comprises: applying the feature information as input to the plurality of reference classification models, and The plurality of reference estimated classification results output from the plurality of reference classification models are obtained.

9. The apparatus according to claim 5, wherein the intermediate information comprises: a first number of votes for a model of a first channel category and a second number of votes for a model of a second channel category by the classification model based on the channel measurement information, and the estimated classification result is determined based on the first number and the second number; and Wherein determining the uncertainty level of the estimated classification result comprises: determining a degree of difference between the first number of the first channel category and the second number of the second channel category, and The level of uncertainty is determined based on the degree of difference.

10. The device according to any one of claims 1 to 9, wherein obtaining the true value classification result comprises: causing a positioning reference device to perform classification marking on the channel measurement information at a location associated with the communication channel; as well as The ground truth classification result is received from the positioning reference device.

11. The apparatus according to any one of claims 1 to 10, wherein the estimated classification result indicates whether the communication channel is classified as a line-of-sight channel or a non-line-of-sight channel.

12. The device according to any one of claims 1 to 11, wherein the device comprises a terminal device or a location management function.

13. A method comprising: determining an importance level of channel measurement information associated with a communication channel for updating a classification model, the classification model being configured to determine an estimated classification result of the communication channel based on the channel measurement information; If it is determined that the importance level of the channel measurement information exceeds the importance threshold of the classification model, obtaining a true value classification result for the communication channel; as well as The classification model is caused to be updated based at least in part on the channel measurement information and the ground-truth classification result.

14. The method according to claim 13, further comprising: determining a level of accuracy of the updated classification model; as well as The importance threshold is updated based on the accuracy level to obtain an updated importance threshold for the updated classification model.

15. The method of claim 14, wherein updating the importance threshold based on the accuracy level comprises: If it is determined that the accuracy level of the updated classification model is higher than the accuracy level of the classification model, the importance threshold is increased by a predetermined value.

16. The method according to any one of claims 13 to 15, wherein determining the importance level comprises: acquiring intermediate information extracted by the classification model from the channel measurement information; as well as The importance level of the channel measurement information is determined based on the intermediate information.

17. The method of claim 16, wherein determining the importance level based on the intermediate information comprises: determining an uncertainty level of the estimated classification result based on the intermediate information; If it is determined that the uncertainty level is the first uncertainty level, determining the importance level of the channel measurement information to be the first importance level; as well as If it is determined that the uncertainty level is a second uncertainty level higher than the first uncertainty level, the importance level of the channel measurement information is determined to be a second importance level, and the second importance level is higher than the first importance level.

18. The method of claim 17, wherein determining the level of uncertainty of the estimated classification result comprises: If it is determined that the classification model is a deep neural network, generating a plurality of reference classification models by applying random neural connection dropout to the classification model; using the plurality of reference classification models, determining a plurality of reference estimated classification results based on the intermediate information; as well as The uncertainty level is determined based on a variance of the plurality of reference classification results.

19. The method of claim 18, wherein the classification model comprises a first model portion and a second model portion connected to the first model portion, and wherein generating the plurality of reference classification models comprises: The plurality of reference classification models are generated by applying random neural connection dropout to the second model portion in the classification model.

20. The method of claim 19, wherein the intermediate information comprises feature information extracted from the channel measurement information by the first model portion of the classification model; and Wherein determining the plurality of reference estimation classification results comprises: applying the feature information as input to the plurality of reference classification models, and The plurality of reference estimated classification results output from the plurality of reference classification models are obtained.

21. The method according to claim 17, wherein the intermediate information comprises: a first number of votes for a model of a first channel category and a second number of votes for a model of a second channel category by the classification model based on the channel measurement information, and the estimated classification result is determined based on the first number and the second number; and Wherein determining the uncertainty level of the estimated classification result comprises: determining a degree of difference between the first number of the first channel category and the second number of the second channel category, and The level of uncertainty is determined based on the degree of difference.

22. The method according to any one of claims 13 to 21, wherein obtaining the true value classification result comprises: causing a positioning reference device to perform classification marking on the channel measurement information at a location associated with the communication channel; as well as The ground truth classification result is received from the positioning reference device.

23. The method according to any one of claims 13 to 22, wherein the estimated classification result indicates whether the communication channel is classified as a line-of-sight channel or a non-line-of-sight channel.

24. An apparatus comprising: means for determining an importance level of channel measurement information associated with a communication channel for updating a classification model, the classification model being configured to determine an estimated classification result of the communication channel based on the channel measurement information; means for obtaining a true classification result for the communication channel if it is determined that the importance level of the channel measurement information exceeds the importance threshold of the classification model; as well as means for causing the classification model to be updated based at least in part on the channel measurement information and the ground-truth classification result.

25. A computer-readable medium comprising instructions stored thereon, the instructions being configured to cause an apparatus to at least perform the method according to any one of claims 13 to 23.

Images