Method, apparatus, and computer program for user equipment positioning

Abstract

A method is provided that is used to determine a first location of a communication device in a first stage, input the first location from the first stage into a machine learning model in a second stage, and determine a second location of the communication device in the second stage at least based on the first location from the first stage by using the machine learning model. The first location is determined in the first stage by using at least one of a non-machine learning model and a machine learning model.

Description

Technical Field

[0001] Various example embodiments relate to the positioning of user equipment. Background Art

[0002] A communication system can be regarded as a facility that enables a communication session between two or more entities (such as communication devices, base stations / access points, and / or other nodes) by providing a carrier wave between various entities involved in a communication path. For example, a communication system can be provided through a communication network and one or more compatible communication devices.

[0003] Access to the communication system can be carried out via a suitable communication device or terminal. The communication device is provided with suitable signal receiving and transmitting means for enabling communication, such as enabling access to a communication network or directly communicating with other communication devices. The communication device can access the carrier wave provided by a station or access point and transmit and / or receive communications on that carrier wave.

[0004] Communication systems and related devices generally operate according to a given standard or specification that defines what the various entities associated with the system are allowed to do and how it should be implemented. Summary of the Invention

[0005] According to one aspect, there is provided a method that includes: determining a first position of a communication device in a first stage; inputting the first position from the first stage into a machine learning model of a second stage; and determining a second position of the communication device in the second stage at least based on the first position from the first stage by using the machine learning model.

[0006] The first position can be determined in the first stage by using at least one of a non-machine learning model and a machine learning model.

[0007] After determining the first position, the first position from the first stage can be directly input into the machine learning model of the second stage.

[0008] The method may further include: determining a first positioning accuracy of the first position in the first stage; and comparing the first positioning accuracy with a second positioning accuracy of the positioning service quality, wherein if the result of comparing the first positioning accuracy and the second positioning accuracy is that the first positioning accuracy is lower than the second positioning accuracy, the first position from the first stage can be input into the machine learning model of the second stage, and wherein the method may further include: if the result of comparing the first positioning accuracy and the second positioning accuracy is that the first positioning accuracy is lower than the second positioning accuracy, inputting the first positioning accuracy from the first stage into the machine learning model of the second stage, and the second position is further determined in the second stage based on the input first positioning accuracy.

[0009] Determining the first location accuracy may include using a look-up table.

[0010] The machine learning model of the second stage may include an autoencoder, and the first location from the first stage may be input into the latent layer of the autoencoder.

[0011] The first positioning accuracy from the first stage may be input into the latent layer of the autoencoder.

[0012] The autoencoder may be trained offline by: calculating a first loss at the output of the autoencoder; calculating a second loss in the latent layer of the autoencoder by using at least the first location of the communication device and the second location of the communication device; and training the autoencoder by using the first loss and the second loss.

[0013] The second loss may include the sum of a first term and a second term, and the second term places the second location within the space around the first location.

[0014] The second term may be based on the maximum absolute distance between the first location and the second location.

[0015] According to one aspect, a method is provided, the method including: receiving, at a machine learning model of a second stage, a first location of a communication device, the first location being determined in a first stage; and determining, at the second stage, a second location of the communication device based at least on the first location from the first stage by using the machine learning model.

[0016] After determining the first location, the first location from the first stage may be directly received at the machine learning model of the second stage.

[0017] If the first positioning accuracy of the first location from the first stage is lower than a second positioning accuracy of a positioning service quality, the first location from the first stage may be received at the machine learning model of the second stage, and the method may further include: receiving, at the second stage, the first positioning accuracy from the first stage if the first positioning accuracy is lower than the second positioning accuracy, and determining the second location at the second stage further based on the received first positioning accuracy.

[0018] The machine learning model of the second stage may include an autoencoder, and the first location from the first stage may be received in the latent layer of the autoencoder.

[0019] The first positioning accuracy from the first stage may be received in the latent layer of the autoencoder.

[0020] The method may further include offline training an autoencoder by: calculating a first loss at an output of the autoencoder; calculating a second loss at a latent layer of the autoencoder by at least using a first location of the communication device and a second location of the communication device; and training the autoencoder using the first loss and the second loss.

[0021] The second loss may include a sum of a first term and a second term, and the second term places the second location within a space around the first location.

[0022] The second term may be based on a maximum absolute distance between the first location and the second location.

[0023] According to one aspect, there is provided an apparatus including: at least one processor; and at least one memory including computer program code, wherein the at least one memory and the computer program code are configured to, with the at least one processor, cause the apparatus to at least: determine a first location of a communication device in a first stage; input the first location from the first stage into a machine learning model in a second stage; and determine a second location of the communication device in the second stage at least based on the first location from the first stage by using the machine learning model.

[0024] The first location may be determined in the first stage by using at least one of a non-machine learning model and a machine learning model.

[0025] After determining the first location, the first location from the first stage may be directly input into the machine learning model in the second stage.

[0026] The at least one memory and the computer program code may further be configured to, with the at least one processor, cause the apparatus to at least: determine a first positioning accuracy of the first location in the first stage; and compare the first positioning accuracy with a second positioning accuracy of a positioning service quality, wherein if a result of a comparison between the first positioning accuracy and the second positioning accuracy is that the first positioning accuracy is lower than the second positioning accuracy, the first location from the first stage may be input into the machine learning model in the second stage, and wherein the at least one memory and the computer program code may further be configured to, with the at least one processor, cause the apparatus to at least: if a result of a comparison between the first positioning accuracy and the second positioning accuracy is that the first positioning accuracy is lower than the second positioning accuracy, input the first positioning accuracy from the first stage into the machine learning model in the second stage, and the second location is further determined in the second stage based on the input first positioning accuracy.

[0027] The first positioning accuracy may be determined by causing the apparatus to at least use a look-up table.

[0028] The machine learning model of the second stage may include an autoencoder, and the first location from the first stage may be input into the latent layer of the autoencoder.

[0029] The first positioning accuracy may be input into the latent layer of the autoencoder.

[0030] At least one memory and computer program code may also be configured to, together with at least one processor, cause the device to offline train the autoencoder at least by: calculating a first loss at the output of the autoencoder; calculating a second loss in the latent layer of the autoencoder by at least using the first location of the communication device and the second location of the communication device; and using the first loss and the second loss to train the autoencoder.

[0031] The second loss may include the sum of a first term and a second term, and the second term causes the second location to be within the space around the first location.

[0032] The second term may be based on the maximum absolute distance between the first location and the second location.

[0033] According to one aspect, there is provided a device, the device comprising: at least one processor; and at least one memory including computer program code, wherein at least one memory and computer program code are configured to, together with at least one processor, cause the device to at least: receive a first location of a communication device at a machine learning model of a second stage, the first location being determined in a first stage; and determine a second location of the communication device at the second stage at least based on the first location from the first stage by using the machine learning model.

[0034] After determining the first location, the first location from the first stage may be directly received at the machine learning model of the second stage.

[0035] If the first positioning accuracy of the first location from the first stage is lower than the second positioning accuracy of the positioning service quality, the first location from the first stage may be received at the machine learning model of the second stage, and at least one memory and computer program code may also be configured to, together with at least one processor, cause the device to at least: if the first positioning accuracy is lower than the second positioning accuracy, receive the first positioning accuracy from the first stage at the second stage, and the second location is further determined at the second stage based on the received first positioning accuracy.

[0036] The machine learning model of the second stage may include an autoencoder, and the first location from the first stage may be received in the latent layer of the autoencoder.

[0037] The first positioning accuracy from the first stage may be received in the latent layer of the autoencoder.

[0038] At least one memory and computer program code may also be configured to, together with at least one processor, cause the apparatus to offline train an autoencoder at least by: calculating a first loss at an output of the autoencoder; calculating a second loss in a latent layer of the autoencoder by at least using a first position of a communication device and a second position of the communication device; and using the first loss and the second loss to train the autoencoder.

[0039] The second loss may include a sum of a first term and a second term, and the second term causes the second position to be within a space around the first position.

[0040] The second term may be based on a maximum absolute distance between the first position and the second position.

[0041] According to one aspect, there is provided an apparatus including components for: determining a first position of a communication device in a first phase; inputting the first position from the first phase into a machine learning model of a second phase; and determining a second position of the communication device in the second phase at least based on the first position from the first phase by using the machine learning model.

[0042] The first position may be determined in the first phase by using at least one of a non-machine learning model and a machine learning model.

[0043] After determining the first position, the first position from the first phase may be directly input into the machine learning model of the second phase.

[0044] The apparatus may further include components for: determining a first positioning accuracy of the first position in the first phase; and comparing the first positioning accuracy with a second positioning accuracy of a positioning service quality, wherein if a result of comparing the first positioning accuracy and the second positioning accuracy is that the first positioning accuracy is lower than the second positioning accuracy, the first position from the first phase may be input into the machine learning model of the second phase, and wherein the apparatus further includes components for: if a result of comparing the first positioning accuracy and the second positioning accuracy is that the first positioning accuracy is lower than the second positioning accuracy, inputting the first positioning accuracy from the first phase into the machine learning model of the second phase, and the second position is further determined in the second phase based on the input first positioning accuracy.

[0045] Determining the first position accuracy may include using a look-up table.

[0046] The machine learning model of the second phase may include an autoencoder, and the first position from the first phase may be input into a latent layer of the autoencoder.

[0047] The first positioning accuracy from the first stage can be input into the latent layer of the autoencoder.

[0048] The apparatus may further include components for offline training of the autoencoder by: calculating a first loss at the output of the autoencoder; calculating a second loss in the latent layer of the autoencoder by using at least a first position of the communication device and a second position of the communication device; and training the autoencoder using the first loss and the second loss.

[0049] The second loss may include the sum of a first term and a second term, and the second term places the second position within a space around the first position.

[0050] The second term may be based on the maximum absolute distance between the first position and the second position.

[0051] According to one aspect, there is provided an apparatus including components for: receiving, at a machine learning model in a second stage, a first position of a communication device, the first position being determined in a first stage; and determining, in the second stage, a second position of the communication device based at least on the first position from the first stage by using the machine learning model.

[0052] After determining the first position, the first position from the first stage may be directly received at the machine learning model in the second stage.

[0053] If the first positioning accuracy of the first position from the first stage is lower than a second positioning accuracy of a positioning service quality, the first position from the first stage may be received at the machine learning model in the second stage; and, and the apparatus may further include components for: receiving, in the second stage, the first positioning accuracy from the first stage if the first positioning accuracy is lower than the second positioning accuracy, and the second position is further determined in the second stage based on the received first positioning accuracy.

[0054] The machine learning model in the second stage may include an autoencoder, and the first position from the first stage may be received in the latent layer of the autoencoder.

[0055] The first positioning accuracy from the first stage may be received in the latent layer of the autoencoder.

[0056] The apparatus may further include components for offline training of the autoencoder by: calculating a first loss at the output of the autoencoder; calculating a second loss in the latent layer of the autoencoder by using at least a first position of the communication device and a second position of the communication device; and training the autoencoder using the first loss and the second loss.

[0057] The second loss may include the sum of a first term and a second term, the second term placing a second location within a space around the first location.

[0058] The second term may be based on a maximum absolute distance between the first location and the second location.

[0059] According to one aspect, there is provided a computer program comprising computer-executable code which, when run on at least one processor, can be configured to cause a device to at least: determine a first location of a communication device in a first stage; input the first location from the first stage into a machine learning model of a second stage; and determine a second location of the communication device in the second stage at least based on the first location from the first stage by using the machine learning model.

[0060] According to one aspect, there is provided a computer program comprising computer-executable code which, when run on at least one processor, can be configured to cause a device to at least: receive, at a machine learning model of a second stage, a location of a first communication device, the first location having been determined in a first stage; and determine a second location of the communication device in the second stage at least based on the first location from the first stage by using the machine learning model.

[0061] According to one aspect, there is provided a computer program comprising computer-executable code which, when run on at least one processor, can be configured to cause a device to perform any of the above methods.

[0062] According to one aspect, there is provided a computer-readable medium having program instructions stored thereon for: determining a first location of a communication device in a first stage; inputting the first location from the first stage into a machine learning model of a second stage; and determining a second location of the communication device in the second stage at least based on the first location from the first stage by using the machine learning model.

[0063] According to one aspect, there is provided a computer-readable medium having program instructions stored thereon for: receiving, at a machine learning model of a second stage, a location of a first communication device, the first location having been determined in a first stage; and determining a second location of the communication device in the second stage at least based on the first location from the first stage by using the machine learning model.

[0064] According to one aspect, there is provided a computer-readable medium having program instructions stored thereon for performing any of the above methods.

[0065] According to one aspect, there is provided a non-transitory computer-readable medium including program instructions stored thereon, the program instructions for: determining a first location of a communication device in a first stage; inputting the first location from the first stage into a machine learning model in a second stage; and determining a second location of the communication device in the second stage at least based on the first location from the first stage by using the machine learning model.

[0066] According to one aspect, there is provided a non-transitory computer-readable medium including program instructions stored thereon, the program instructions for: receiving, at a machine learning model in a second stage, a location of a first communication device, the first location being determined in a first stage; and determining a second location of the communication device in the second stage at least based on the first location from the first stage by using the machine learning model.

[0067] According to one aspect, there is provided a non-transitory computer-readable medium including program instructions stored thereon, the program instructions for performing any of the above methods.

[0068] According to one aspect, there is provided a non-volatile tangible storage medium including program instructions stored thereon, the program instructions for: determining a first location of a communication device in a first stage; inputting the first location from the first stage into a machine learning model in a second stage; and determining a second location of the communication device in the second stage at least based on the first location from the first stage by using the machine learning model.

[0069] According to one aspect, there is provided a non-volatile tangible storage medium including program instructions stored thereon, the program instructions for: receiving, at a machine learning model in a second stage, a location of a first communication device, the first location being determined in a first stage; and determining a second location of the communication device in the second stage at least based on the first location from the first stage by using the machine learning model.

[0070] According to one aspect, there is provided a non-volatile tangible storage medium including program instructions stored thereon, the program instructions for performing any of the above methods.

[0071] According to one aspect, there is provided an apparatus including circuitry configured to: determine a first location of a communication device in a first stage; input the first location from the first stage into a machine learning model in a second stage; and determine a second location of the communication device in the second stage at least based on the first location from the first stage by using the machine learning model.

[0072] According to one aspect, a device is provided that includes circuitry configured to: receive a location of a first communication device at a machine learning model in a second phase, the first location being determined in a first phase; and determine a second location of the communication device in the second phase based at least on the first location from the first phase by using the machine learning model.

[0073] According to one aspect, a device is provided that includes circuitry configured to perform any of the methods described above.

[0074] In the foregoing, a number of different aspects have been described. It should be understood that other aspects may be provided by any combination of two or more of the above aspects.

[0075] Various other aspects are also described in the following detailed description and the appended claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Some example embodiments will now be described with reference to the following drawings:

[0077] Figure 1 An example embodiment of a wireless communication system is shown;

[0078] Figure 2 An example embodiment of a communication device is shown;

[0079] Figure 3 An example embodiment of a non-transitory computer-readable medium is shown;

[0080] Figure 4 An example embodiment of a device is shown;

[0081] Figure 5 An example embodiment of a non-roaming architecture is shown;

[0082] Figure 6 An example graph showing the expected horizontal positioning accuracy of different positioning methods in different scenarios is shown;

[0083] Figure 7 An example embodiment of a positioning process is schematically shown;

[0084] Figure 8 An example embodiment of a comparison level of positioning accuracy provided by different positioning methods of a positioning process is schematically shown;

[0085] Figure 9 An example embodiment of a time scale representation in the case where a UE follows a trajectory is shown;

[0086] Figure 10 An example implementation of a positioning process is shown;

[0087] Figure 11 Shows an example embodiment of an autoencoder;

[0088] Figure 12 Shows an example flowchart depicting an example embodiment of a positioning process;

[0089] Figure 13 Shows an example flowchart depicting an example embodiment of a positioning process performed by an ML model of an ML-based positioning method;

[0090] Figure 14 Shows an example embodiment of an environment with a UE location; and

[0091] Figure 15 Shows an example embodiment of the cumulative density function approximation of the estimation errors of three positioning methods. DETAILED DESCRIPTION

[0092] As used herein, it should be noted that unless otherwise specified, the terms "positioning", "location", and "place" and their derivatives or variants will be used with a similar meaning.

[0093] As is well known, a wireless system can be divided into cells and is thus commonly referred to as a cellular system. Generally, an access point such as a base station provides at least one cell. A cellular system can support communication between user equipment (UE) devices. This application relates to cellular radio implementations, including 2G, 3G, 4G, and 5G radio access networks (RANs); cellular Internet of Things (IoT) RANs; and cellular radio hardware.

[0094] Hereinafter, certain embodiments are explained with reference to communication devices capable of communicating via a wireless cellular system and a communication system serving such communication devices. Before explaining the example embodiments in detail, reference is made to Figures 1 to 4 Briefly explain certain general principles of a wireless communication system, its access system, and communication devices to assist in understanding the technology underlying the examples described.

[0095] An example of a wireless communication system is an architecture standardized by the Third Generation Partnership Project (3GPP). The latest 3GPP-based development is commonly referred to as 5G or New Radio (NR). Other examples of radio access systems include radio access systems provided by base stations of systems based on technologies such as Wireless Local Area Network (WLAN) and / or Worldwide Interoperability for Microwave Access (WiMAX). It should be understood that although some embodiments are described in the context of a 5G system, other embodiments can be provided in any other suitable system, including but not limited to subsequent systems or similar protocols defined outside the 3GPP forum.

[0096] Figure 1An example embodiment of a wireless communication system 100 is shown. As can be seen therein, a communication device 102 is served by a first cell 104, which is provided by a first base station 108a.

[0097] In this example embodiment, the communication device 102 may be served by a second cell 106, which is provided by a second base station 108b. Depending on the system 100, the base stations 108a, 108b may be any suitable transmission and reception point (TRP). For example, a TRP (sometimes referred to as a TRxP) may be a gNB or an ng-eNB (the term "ng" means "next generation").

[0098] Hereinafter, the communication device will be referred to as a UE. However, it should be understood that the device may be any suitable communication device, and the term UE may be intended to cover any such device. Some examples of communication devices are discussed below, and as used in this document, the term UE is intended to cover any one or more of these devices and / or any other suitable device. The communication device has a wireless connection to the base station.

[0099] Figure 2 An example embodiment of a communication device 200 such as Figure 1 the communication device 102 is shown.

[0100] The communication device 200 may be provided by any device capable of transmitting and receiving radio signals. Non-limiting examples include a mobile station (MS) or a mobile device (such as a mobile phone or a so-called "smartphone"), a computer with a wireless interface card or other wireless interface facilities (e.g., a USB dongle), a personal data assistant (PDA) or a tablet computer with wireless communication capabilities, a machine type communication (MTC) device, an IoT type communication device, or any combination thereof, etc.

[0101] As can be seen, the communication device 200 may include a transceiver device 210, which is configured to receive and transmit signals via an air or radio interface 212. For example, the transceiver device 210 may be provided by a radio part and an associated antenna arrangement. In addition, the antenna arrangement may be disposed inside or outside the communication device 200.

[0102] The communication device 200 may include at least one processor 202 and at least one memory 204. The at least one memory 204 may include at least one read-only memory (ROM) and / or at least one random access memory (RAM). The communication device 200 may include other possible components 206 for use in the software and hardware assisted execution of the tasks that the communication device 200 is designed to perform, including control of access to and communication with access systems and other communication devices. The at least one processor 202 is coupled to the at least one memory 204. The at least one processor 202 may be configured to execute appropriate software code to implement one or more of the following embodiments. The software code may be stored in the at least one memory 204, for example, stored in the at least one ROM.

[0103] The processor 202, storage device, and other associated control devices may be provided on a suitable circuit board and / or in a chipset (represented by reference numeral 208).

[0104] The communication device 200 may optionally have a user interface, such as a keypad, a touch-sensitive screen or keyboard, or any combination thereof, etc.

[0105] Optionally, depending on the type of the communication device 200, one or more of a display, a speaker, and a microphone may be provided.

[0106] Generally, communication protocols and / or parameters applied to the connection are also defined. The communication device 200 may access the communication system 100 based on various access technologies.

[0107] Figure 3 Example embodiments of non-transitory computer-readable media 300a (e.g., a computer disk (CD) or a digital versatile disk (DVD)) and 300b (e.g., a universal serial bus (USB) memory stick) are shown. The non-transitory computer-readable media 300a and 300b are configured to store instructions and / or parameters 302 that, when executed by a processor, allow the processor to execute one or more steps of any method of any embodiment.

[0108] Figure 4 An example embodiment of a device 400 is shown.

[0109] The device 400 may be provided in any network entity. The device 400 may include at least one processor 410 and at least one memory 420, and the at least one memory 420 includes computer code for one or more programs. The device 400 may be configured to cause some embodiments to be executed and provided in, for example, a location management function (LMF).

[0110] One or more of the following aspects relate to a 5G system (5GS). The new radio interface introduced as part of the 5GS is called New Radio (NR). However, it should be understood that some of these aspects can be used with any other suitable radio access technology system, such as Long Term Evolution (LTE) of Universal Mobile Telecommunications System (UMTS), UMTS Terrestrial Radio Access Network (UTRAN) (3G radio), and / or any other suitable system.

[0111] Figure 5 A non-roaming architecture 500 for a positioning service is shown in a service-based interface representation, where in addition to the communication device and its interfaces with the network, the network can also provide various other functions.

[0112] Some of these network functions can be provided by the Core Network (CN). Example embodiments of the present application can be provided by such a network that provides these functions. Similarly, although the present specification refers to the UE here and elsewhere, those skilled in the art should understand that the communication device can be another type of communication device different from the UE.

[0113] The LMF 510 of the architecture 500 is a network entity in the 5G CN (5GC) that is responsible for supporting the location determination of the UE 520, obtaining downlink (DL) location measurements or location estimates from the UE 520, obtaining uplink (UL) location measurements from the Radio Access Network (NG-RAN) 550, and obtaining non-UE related assistance data from the NG-RAN 550. In some example embodiments, the LMF 510 can expose its services to external and / or internal clients using service-based application programming interfaces (APIs).

[0114] The Unified Data Management (UDM) function 530 of the architecture 500 stores subscription information and supports the Authentication Credential Repository and Processing Function (ARPF) and stores security credentials for authentication.

[0115] The Access and Mobility Management Function (AMF) 540 of the architecture 500 is configured to perform multiple tasks, including: registration management, connection management, reachability management, mobility management, and various functions related to security and access management and authorization. The AMF 540 provides these services to devices such as the UE 520, which is configured to communicate with the CN.

[0116] The NG-RAN 550 of the architecture 500 is configured to provide access to the CN for the UE 520. The NG-RAN 550 includes one or more base stations and one or more associated Radio Network Controllers (RNCs).

[0117] The Network Data Analytics Function (NWDAF) 560 of architecture 500 is configured to perform data analysis based on requests from one or more network functions in the 5G network. In some example embodiments, the NWDAF 560 may expose its services to external and / or internal clients using an API.

[0118] The Location Service Client (LCS) 570 of architecture 500 is configured to send and receive communications to and from the CN as an external client.

[0119] The Gateway Mobile Location Center (GMLC) 580 of architecture 500 includes the functions needed to support location-based services and thus interfaces with external clients such as the LCS 570.

[0120] The Location Retrieval Function (LRF) 590 of architecture 500 can be used to retrieve the location information of a user who initiates an emergency session.

[0121] UE location information can be used by the network for radio resource management (RRM) and also for providing critical emergency services.

[0122] In architecture 500, reference points for supporting location services are also provided, such as: "N2" between the NG-RAN 550 and the AMF 540; and "Le" between the GMLC 580 and the LCS client 570 and between the LRF 590 and the LCS client 570.

[0123] In architecture 500, service-based interfaces for supporting location services are also provided, such as: "Nlmf" shown by the LMF 510; "Nudm" shown by the UDM 530; "Namf" shown by the AMF 540; "Nnwdaf" shown by the NWDAF 560; and "Ngmlc" shown by the GMLC 580.

[0124] In this regard, several methods for performing UE location based on non-machine learning (non-ML) and machine learning (ML) have been proposed.

[0125] For example, some of these non-ML-based location methods can be found in: "Survey of cellular mobile radio localization methods: from 1G to 5G" by J.A. del Peral-Rosado et al., IEEE Communications Surveys & Tutorials, vol. 20, no. 2, pp. 1124-1148, second quarter 2018. Among them, the following can be cited:

[0126] · Trilateration: The UE position is obtained by calculating the intersection between geometric shapes (e.g., circles or hyperbolas) created from distance measurements between the terminal and reference transmitters or receivers. Examples of measurements can be Time of Arrival (ToA), Time Difference of Arrival (TDoA), or Received Signal Strength (RSS);

[0127] · Triangulation: Utilizes the direction of the received signal or Angle of Arrival (DoA or AoA) to estimate the position using the intersection of at least two known directions of the incoming signal;

[0128] · Proximity: Designates the position of a known transmitter as the position of the terminal. An example can be the Cell ID (CID)-based method, where the provided position is one of the serving base stations;

[0129] · Scene analysis: Also known as fingerprinting or pattern matching, the algorithm is based on finding the best match in a fingerprint database for a certain signal measurement, such as received signal power, time delay, or channel delay spread, where each fingerprint is associated with a specific location; and

[0130] · Hybrid: Can implement combinations of previous localization algorithms to improve overall performance or support algorithms that cannot be calculated independently due to lack of signal measurements.

[0131] UE localization can be identified as a prerequisite for emergency reasons as well as for other purposes, such as network optimization and Location-Based Services (LBS). So far, non-ML-based localization methods have provided UE position accuracy in response to requirements set by regulatory bodies for emergency purposes. For example, regulatory bodies such as the Federal Communications Commission (FCC) in the United States have defined Enhanced 911 (E911) localization requirements that specify a horizontal localization accuracy of 50 meters and a vertical localization accuracy of 3 meters for 67% of 911 calls.

[0132] Retrieved from: “Survey of cellular mobile radiolocalization methods: from 1G to 5G” by J.A. del Peral-Rosado et al., IEEE Communications Surveys & Tutorials, vol. 20, no. 2, pp. 1124 - 1148, Second Quarter 2018, Figure 6Shows the expected horizontal accuracy of cellular mobile radio localization methods, such as cell ID, radio frequency pattern matching (RFPM), TDoA, assisted global navigation satellite system (A-GNSS), and hybrid, which are applicable to different scenarios such as indoor, outdoor urban, and rural. It can be seen that the methods defined for 2G / 3G / 4G networks can achieve an optimal accuracy of 1 meter through WLAN / Bluetooth measurements.

[0133] Also retrieved from: "Survey of cellular mobile radiolocalization methods: from 1G to 5G" by J.A. del Peral-Rosado et al., IEEE Communications Surveys & Tutorials, vol. 20, no. 2, pp. 1124-1148, Second Quarter 2018, Table I below provides a description of the localization accuracies achievable by different localization methods in horizontal and vertical planes, such as cell ID + timing advance (CID + TA), cell ID + round-trip time (CID + RTT), enhanced cell ID (E-CID), RFPM, UL-ToA, UL-TDoA (UTDoA), enhanced OTD (E-OTD), advanced forward link trilateration (AFLT), observed TDoA (OTDoA), A-GNSS, terrestrial beacon system (TBS), barometer, and hybrid, and their compatible radio access technologies (RAT), such as 2G, 3G, 4G, WLAN, and Bluetooth.

[0134] Table I: Classification of standardized localization methods in cellular systems

[0135]

[0136] Nonetheless, 5G networks specify new use cases and scenarios that require higher positioning accuracy compared to what has been achieved so far with the positioning accuracy and standardized NR positioning methods stipulated by regulatory bodies. For example, as specified in 3GPP TR 22.862 v14.1.0, high positioning accuracy includes the following requirements: location information is obtained quickly, reliably, and is available (e.g., the location can be determined). For example, a typical area that requires "higher accuracy positioning" could be vehicle collision avoidance: each vehicle must know its own location, the locations of nearby vehicles, and their expected paths to avoid collisions. Next-generation high-accuracy positioning requires an accuracy of less than 1 meter in more than 95% of the service area (including indoor, outdoor, and urban environments). Specifically, network-based positioning in three-dimensional space should support an accuracy of 10 meters to <1 meter in 80% of cases, and the accuracy of indoor deployments should be better than 1 meter. High-accuracy positioning services in 5G networks should be supported in areas such as traffic roads, tunnels, underground parking lots, or indoor environments.

[0137] In this regard, ML-based positioning methods can have the advantage of higher positioning accuracy compared to non-ML positioning methods. Example ML-based positioning methods can be based on models such as artificial neural networks (ANNs), including, for example, deep neural networks (DNNs) and convolutional neural networks (CNNs), decision trees, support vector machines (SVMs), regression analysis, Bayesian networks, and genetic algorithms (GAs). Example ML-based positioning methods can be found in "An efficient machine learning approach for indoor localization" by L. Zhang et al., China Communications, vol. 14, no. 11, pp. 141-150, November 2017. In which the ML method combines a grid-search-based kernel support vector machine and principal component analysis, applies principal component analysis to reduce high-dimensional measurements, and designs a grid-search algorithm to optimize the parameters of the kernel support vector machine to improve positioning accuracy. The experimental results show that the proposed ML method reduces the positioning error and improves the computational efficiency of methods based on K-nearest neighbor, backpropagation neural network, and support vector machine.

[0138] Figure 7 An exemplary embodiment of the positioning process 700 is schematically shown, where the reference numeral "SW" depicted schematically refers to a two-position switch (two positions are depicted as ① and ②) to help better schematically visualize two exemplary embodiments according to which the positioning process is performed.

[0139] In the first example stage, a first position of a communication device (e.g., a UE) is determined, by Indication. Although Figure 7 the example embodiments of Figure 7 show that the determination of the first position is performed by a non-ML-based positioning method 710 (described as a non-ML method) using a non-ML model, it should be noted that in another example embodiment, the determination of the first position can be performed by an ML-based positioning method using an ML model or any other suitable positioning method. The determination of the first position includes estimating the first position of the communication device such that the determined first position is the estimated position of the communication device.

[0140] The non-ML-based positioning method 710 can be selected from any positioning method defined by 3GPP, such as but not limited to any of the positioning methods listed in Table I above. In this regard, the selection of the non-ML-based positioning method 710 can depend on the RAT under consideration and the available inputs in the network that can be used to estimate UE positioning, such as Wi-Fi measurements, positioning reference signal (PRS) measurements, sounding reference signal (SRS) measurements, etc. As shown, the non-ML model of the non-ML-based positioning method 710 receives a first set of measurements (depicted as Measurement Set 1) as input and provides, based on the received input, a first positioning result (i.e., the first position of the communication device), represented by as output.

[0141] In a first example embodiment (as Figure 7 shown, schematically corresponding to SW being switched to position ①), the determination of the first positioning accuracy of the first position (represented by 720) can be performed, and the first positioning accuracy can be compared (represented by 730) with the target positioning accuracy of the positioning quality of service (QoS) (represented by 740) (depicted as the second positioning accuracy). The determination of the first positioning accuracy (represented by 720) includes estimating the positioning accuracy of the first position such that the determined first positioning accuracy is the estimated positioning accuracy of the first position The first positioning accuracy is a positive number, which can be represented in any suitable manner, for example, as a percentage value or a mean square value. For example, the first positioning accuracy can be determined by using a look-up table (LUT) with a dedicated database, which is configured to provide a corresponding average positioning accuracy for each of a plurality of non-ML-based positioning methods and thus for the application used to determine the first position ​The non-ML based positioning method 710 provides a corresponding average positioning accuracy; or a first positioning accuracy It can be determined in an offline manner based on the selected non-ML based positioning method 710 and its parameters while considering the propagation environment such as indoor and outdoor. The target positioning accuracy (i.e., the second positioning accuracy) can be, for example, the required positioning accuracy included in the QoS structure defined by 3GPP Release 16.

[0142] If the comparison result is the first positioning accuracy If the positioning accuracy is equal to or greater than the second positioning accuracy, the non-ML based positioning method 710 can be considered to determine the first position The accuracy is satisfactory, and the positioning process 700 provides a first position (also corresponding to the output of the non-ML model based on the non-ML positioning method 710) as an output (represented by 750). On the other hand, if the result of the comparison is the first positioning accuracy is lower than the second positioning accuracy, which means that the non-ML based positioning method 710 determines the first position If the second positioning accuracy cannot be considered satisfactory with respect to the accuracy of the first positioning accuracy, an additional stage represented by the second exemplary stage is required to improve the positioning accuracy with respect to the first positioning accuracy. The position of the communication device is determined with increased positioning accuracy, wherein the aim is to reach a target positioning accuracy for the QoS, ie a second positioning accuracy for the QoS. The second positioning accuracy may be expressed in any suitable manner, for example as a percentage value or as a mean square value.

[0143] In order to reduce any signaling overhead, for example due to PRS transmission, a second exemplary embodiment may be provided (schematically corresponding to SW being switched to position ②). In the second exemplary embodiment, the first position First position accuracy The determination 720 and the comparison 730 with the second position accuracy of the QoS are not performed in the first example stage, and the first position is directly provided to the second example stage. In this second example embodiment, with respect to the first position The first positioning accuracy is not provided to the second example stage, and information about the target positioning accuracy (ie, the second positioning accuracy of the location QoS) can be directly obtained in the second example stage due to the availability of said information about the target positioning accuracy in the network.

[0144] In the second example stage, if the comparison result is the first positioning accuracy lower than the second positioning accuracy or if the first position is directly provided to the second example stage (i.e., without determining the first location accuracy of the first positioning and comparing the first positioning accuracy with the second positioning accuracy of the location QoS in the step), then the ML-based positioning method 760 (represented by the ML method) using the ML model is applied to determine the second location of the communication device (such as a UE), denoted as second location The determination of includes estimating the second location of the communication device such that the determined second location is the estimated location of the communication device.

[0145] The ML-based positioning method 760 can be based on ANN models such as, for example but not limited to, DNN models, CNN models, etc. As Figure 7 shown, when the SW is set to position ①, if the result of the above comparison is that the first positioning accuracy is lower than the second positioning accuracy, then the ML model of the ML-based positioning method 760 receives the first location of the communication device from the first example stage and the first positioning accuracy as well as the second set of measurements (depicted as measurement set 2) as inputs. On the other hand, when the SW is set to position ②, the ML model of the ML-based positioning method 760 receives the first location of the communication device from the first example stage as well as the second set of measurements (depicted as measurement set 2) as inputs.

[0146] The positioning process 700 provides the second positioning result represented by (which also corresponds to the output of the ML model of the ML-based positioning method 760) (i.e., the second location of the communication device) as an output (represented by 770).

[0147] The positioning process 700 can be based on supervised learning. To use it, offline training of the ML model is necessary, as indicated by the reference numeral 780.

[0148] In an example embodiment, the real-time functionality of the positioning process 700 can be implemented at the LMF 510. In an example option, the training can be run at the NWDAF 560, and then the training data can be collected through minimized drive test (MDT) measurements, and the corresponding measurement logs / reports of the MDT measurements can include: time information, RF measurements, and detailed location information (e.g., GNSS location information). Since the MDT may already be available as an input to the NWDAF 560, it is necessary to transfer the trained ML model to the LMF 510 for online use. In another example option, the training and inference operations can be run at the LMF 510. For this purpose, it is necessary to perform MDT measurements on the LMF side in order to perform the required training of the ML model.

[0149] In an example embodiment in which the ML-based positioning method 760 is designated to run in the NWDAF 560, the LMF 510 can use the API provided by the NWDAF 560.

[0150] Therefore, Figure 7 the positioning process 700 can be based on the hybrid use of a first positioning method (i.e., Figure 7 the non-ML-based positioning method 710 in the example embodiment of Figure 7 and a second positioning method (i.e., the ML-based positioning method 760 in the example embodiment of ). In addition, the positioning process 700 can be a two-stage process, including a first example stage (related to the non-ML-based positioning method 710 and an optional comparison 730 between a first positioning accuracy and a second positioning accuracy), followed by a second example stage (related to the ML-based positioning method 760, where if the result of the above comparison is that the first positioning accuracy Figure 7 is lower than the second positioning accuracy or if the first position

[0151] Figure 8 is directly provided to the second example stage, the output of the first example stage is provided as the input to the second example stage). Therefore,

[0152] the positioning process 700 can be considered a hierarchical hybrid positioning (HHL) process. the "coarse" level of positioning accuracy of the communication device, i.e., the first positioning accuracy After the output of the non-ML model of the non-ML based positioning method 710 has been provided as input to the ML model of the ML based positioning method 760, the ML based positioning method 760, depicted as ML positioning, can schematically provide a first position of the communication device at a "fine" level of positioning accuracy to obtain a second position of the communication device Thus, the ML based positioning method 760 of the positioning process 700 can be schematically regarded as a "zooming" or "refinement" method.

[0153] It should be noted that the first example stage and the second example stage of the positioning process 700 can run at different time scales, depending on the availability of the first set of measurements and the second set of measurements input at the level of each corresponding example stage.

[0154] At this point, Figure 9 an example embodiment of a time scale representation is shown in the case where a communication device such as a UE follows a trajectory within a Cartesian coordinate (x, y) system.

[0155] It can be seen that the non-ML based positioning method 710, depicted as non-ML positioning, provides a greater positioning error compared to the ML based positioning method 760, depicted as ML positioning. Thus, at different times, the ML based positioning method 760 can provide a more accurate UE position within the position range of the non-ML based positioning method 710 compared to the non-ML based positioning method 710.

[0156] Figure 10 An example implementation 1000 of the positioning process 700 is shown, where the non-ML model of the non-ML based positioning method 710 is the UTDoA model of the UTDoA based positioning method 1010 (represented by UTDoA), and the ML model of the ML based positioning method 760 is the DNN model of the DNN based positioning method 1060 (represented by DNN).

[0157] In Figure 10 this example implementation 1000, the first position of the communication device provided by the first example stage is depicted as (which is the estimated position); the second position of the communication device provided by the second example stage is depicted as (which is the estimated position); the first set of measurements, depicted as measurement set 1, can include SRS measurements and corresponding configuration parameters; the second set of measurements, depicted as measurement set 2, can include reference signal received power (RSRP) measurements reported by the communication device (e.g., UE), and UL-TOA measurements. Still in Figure 10 this the first positioning accuracy of the first position The determination (indicated by 1020) of (which is the estimated positioning accuracy) is determined by using a LUT having a database 1080 configured to provide a corresponding average positioning accuracy for a UTDoA-based positioning method; the positioning QoS providing a second positioning accuracy is indicated by 1040; the comparison between the first positioning accuracy and the second positioning accuracy is indicated by 1030.

[0158] In addition, in the case where the result of the comparison is that the first positioning accuracy is equal to or greater than the second positioning accuracy, an example implementation 1000 of the positioning process 700 can provide a first position as an output (indicated by 1050). On the other hand, in the case where the first positioning accuracy is lower than the second positioning accuracy or the first position is directly provided to the second example stage, an example implementation 1000 of the positioning process 700 can provide a second position (which also corresponds to the output of the DNN model of the DNN-based positioning method) as an output (indicated by 1070).

[0159] The DNN model of the DNN-based positioning method 1060 can include a neural network, for example, an autoencoder. In an example embodiment, the autoencoder can be a CNN, DNN, or any other suitable ANN.

[0160] Figure 11 An example embodiment of the autoencoder 1100 is shown. The autoencoder 1100 can include an encoder 1110, a latent space representation 1120 (also referred to as a code), and a decoder 1130.

[0161] The encoder 1110 can include one or more hidden layers and is configured to compress input data (depicted as X RSRP-TOA ) into the latent space representation 1120, which input data can correspond to a second set of measurements including RSRP and UL-TOA measurements. The latent space representation 1120 includes a single hidden layer represented by a latent layer as the most intermediate layer. It can be seen that the latent layer can receive the first position of the communication device from the first example stage of the positioning process 700 (which is the estimated position), and when the first positioning accuracy is lower than the second positioning accuracy, receive the first position from the first example stage of the positioning process 700 of the first positioning accuracy (which is the estimated positioning accuracy), and then can provide the second position of the communication device (which is the estimated position) as output. The decoder 1130 may include one or more hidden layers and is configured to reconstruct the input data from the latent space representation by providing the reconstructed input data (depicted as ) as output. As shown, each hidden layer of the encoder 1110, the latent space representation 1120, and the decoder 1130 has a corresponding number of neurons (also known as nodes), with the latent layer having the fewest neurons.

[0162] The ML model requires offline training and the provision of training data or samples. As an example, the autoencoder 1100 of the ML model can be offline trained according to the following exemplary embodiments.

[0163] Let S be the set of training samples, L be the set of labeled samples (i.e., having available ground truth positions), and U be the set of remaining unlabeled samples. So, S is the union of L and U. For each element a ∈ L, we have the availability of the 4-tuple where X RSRP-TOA，a , y a , and represent the RSRP and UL-TOA measurements (X RSRP-TOA，a ), the actual position of the communication device (y a ), the first position of the communication device (which is the estimated position determined by the UTDoA-based positioning method 1010 using the UTDoA model), and the first positioning accuracy (which is a positive number corresponding to the estimated positioning accuracy of the first position determined by the UTDoA-based positioning method 1010 using the UTDoA model). For each element a ∈ U, we only have the availability of the 3-tuple .

[0164] The input to the encoder 1110 of the autoencoder 1100 is X R5RP-TOA，a , and for all a ∈ S, at the output of the decoder 1130 of the autoencoder 1100, this input is reconstructed as Then, the sum of the first loss at the output of the decoder 1130 of the autoencoder 1100 and the second loss at the hidden layer 1120 of the autoencoder 1100 can be used to train the autoencoder 1100.

[0165] The first loss at the output of the decoder 1130 can also be specified as the reconstruction loss of the autoencoder 1100 and is given by the following relation (1):

[0166]

[0167] Second loss at the latent layer 1120 can be given by the following relation (2):

[0168] It can be seen that the second loss is the sum of a first loss term and a second loss term. The first loss term is the mean squared error (MSE) loss between the actual location of the available labeled data and the estimated location of the communication device (i.e., the second location) determined by the DNN-based positioning method 1070 using the autoencoder 1100. The second loss term is the maximum absolute distance (MAD) loss. It is used to limit the estimated location (i.e., the second location) from the second example phase using the autoencoder 1100 so that it lies within the space around the corresponding estimated location (i.e., the corresponding first location) from the first example phase using the UTDoA model, thereby allowing a higher accuracy to be achieved with a smaller amount of labeled data.

[0169] Therefore, compared with an independent ML model using an ML-based positioning method, using an ML model (e.g., the autoencoder 1100) and the estimated locations from a non-ML model of a non-ML-based positioning method can shorten the offline training phase of the ML model because the ML model requires a smaller amount of labeled data to achieve the same level of positioning accuracy. It also allows for an improvement in positioning accuracy compared to the positioning accuracy provided by an independent non-ML model.

[0170] Figure 12 FIG. 1200 is an example flowchart that illustrates an example embodiment of the positioning process 700. The example flowchart 1200 can be implemented in, for example, the LMF 510 or a device within the LMF 510.

[0171] In step 1210, the method of the example flowchart 1200 can include determining a first location of the communication device in a first phase

[0172] In step 1220, the method of the example flowchart 1200 can include inputting the first location from the first phase into an ML model of a second phase.

[0173] In step 1230, the method of the example flowchart 1200 can include determining a second location of the communication device in the second phase based at least on the first location from the first phase by using the ML model

[0174] Figure 13FIG. 1300 is an example flowchart showing an example embodiment of a positioning process 700 performed by an ML model of an ML-based positioning method 760. The example flowchart 1300 may be implemented in, for example, the LMF 510 or the NWDAF 560 or a device within the LMF 510 or the NWDAF 560.

[0175] In step 1310, the method of the example flowchart 1300 may include receiving, at an ML model in a second phase, a first position of a communication device The first position is determined in a first phase.

[0176] In step 1320, the method of the example flowchart 1300 may include determining, in the second phase, a second position of the communication device based at least on the first position from the first phase by using the ML model

[0177] Figure 14 FIG. shows an example embodiment of an environment with 2000 UE devices placed within a Cartesian coordinate (x, y) system, where x and y are in meters (m).

[0178] For simulation purposes, there are 21 gNBs, each UE reports the strongest beam RSRP to each of the 21 gNBs, and UL-TOA is measured at each gNB for each of the 2000 UE devices. In a first example positioning phase, a non-ML-based positioning method using UTDoA measurements, such as the UTDoA-based positioning method 1010, is applied to obtain the estimated UE positions of all 2000 UE devices. In the next positioning phase (i.e., the second example phase), an ML model (such as Figure 11 the autoencoder 1100) is used for positioning. The encoder part 1110 of the autoencoder 1100 has two dense hidden layers, which respectively have 128 and 32 neurons, the latent layer 1120 of the autoencoder 1100 has two neurons, and the decoder part 1130 of the autoencoder 1100 has two dense hidden layers, which respectively have 32 and 128 neurons. Both the input layer and the output layer have 42 neurons. The input X RSRP-TO-A of the autoencoder is a 42-dimensional vector composed of a 21-dimensional RSRP vector and a 21-dimensional TOA vector. The input (including the estimated position and the first positioning accuracy from the first example positioning phase) is provided to the latent layer 1120. In addition, ground truth labels are provided for 1000 out of the 2000 samples, and the above-mentioned first loss at the output of the decoder part 1130 and the second loss Offline training of the autoencoder 1100.

[0179] Based on the above simulation parameters, Figure 15 An example embodiment of a cumulative density function (CDF) plotted for the estimated error values (in meters) for the following three positioning methods is shown:

[0180] · A positioning method 1010 based on (independent) UTDoA that uses UTDoA measurements as inputs to the UTDoA model;

[0181] · A positioning method 1060 based on DNN that uses the estimated position and the first positioning accuracy provided by the UTDoA model to the autoencoder 1100 (labeled AE) that is the DNN model; and

[0182] · An independent DNN-based positioning method 1060 whose autoencoder (as the DNN model) is trained only on 1000 samples without using the estimated position and the first positioning accuracy from the first example phase.

[0183] Table II below provides a description of the average estimated error (in meters) calculated for each of the above three positioning methods on 1000 unlabeled samples.

[0184] Table II: Average Estimated Error (Meters) of the Above Three Positioning Methods

[0185] Method Average Estimation Error (m) UTDOA 32.93 UTDOA + AE 18.45 Independent DNN 25.12

[0186] Table II shows that the average estimated error of the DNN-based positioning method 1060 that uses AE 1100 and the estimated position and the first positioning accuracy provided by the UTDoA model is less than the average estimated error of the independent DNN-based positioning method and the UTDoA-based positioning method. It should be noted that when offline training is performed on a large number of samples, the accuracy gain of the independent ML-based positioning method is more significant.

[0187] It should be noted that although the above example embodiments, without departing from the scope of the present application, several variations and modifications can be made to the disclosed solutions. For example, although Figures 7 to 14The example embodiments illustratively relate to a positioning process based on a non-ML-based positioning method using a non-ML model in a first example phase and based on an ML-based positioning method using an ML model in a second example phase. However, in another example embodiment, the proposed positioning process may be based on an ML-based positioning method using an ML model in both the first example phase and the second example phase, or based on any other suitable positioning method using a suitable model in the first example phase and based on an ML-based positioning method using an ML model in the second example phase.

[0188] Accordingly, embodiments may vary within the scope of the appended claims. In general, some embodiments may be implemented in hardware or dedicated circuits, software, logic, or any combination thereof. For example, some aspects may be implemented in hardware, while other aspects may be implemented in firmware or software executable by a controller, microprocessor, or other computing device, but the embodiments are not limited thereto. Although various embodiments may be illustrated and described as block diagrams, flowcharts, or using some other graphical representation, it is well understood that, by way of non-limiting example, the blocks, devices, systems, techniques, or methods described herein may be implemented in hardware, software, firmware, dedicated circuits or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.

[0189] Embodiments may be implemented by computer software stored in a memory and executable by at least one data processor of the entities involved, or by hardware, or by a combination of software and hardware. Further, in this regard, it should be noted that any one of the above processes may represent program steps, or interconnected logic circuits, blocks, and functions, or a combination of program steps and logic circuits, blocks, and functions. The software may be stored on a physical medium, such as a memory chip or a memory block implemented within a processor, a magnetic medium such as a hard disk or a floppy disk, and an optical medium such as a DVD and its data variant CD.

[0190] The memory may be of any type suitable for the local technical environment and may be implemented using any suitable data storage technology, such as semiconductor-based memory devices, magnetic memory devices and systems, optical memory devices and systems, fixed memory, and removable memory. The data processor may be of any type suitable for the local technical environment and, by way of non-limiting example, may include one or more of a general-purpose computer, a dedicated computer, a microprocessor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), gate-level circuitry, and a processor based on a multi-core processor architecture.

[0191] Alternatively or additionally, some embodiments may be implemented using circuitry. The circuitry may be configured to perform one or more of the previously described functions and / or method steps. The circuitry may be provided in a base station and / or a communication device.

[0192] As used in this application, the term "circuitry" may refer to one or more or all of the following:

[0193] (a) A pure hardware circuit implementation (such as an implementation using only analog and / or digital circuitry);

[0194] (b) A combination of hardware circuitry and software, for example:

[0195] (i) A combination of analog and / or digital hardware circuitry and software / firmware, and

[0196] (ii) Any part of a hardware processor (including a digital signal processor), software, and memory with software that work together to cause a device (such as a communication device or a base station) to perform various previously described functions; and

[0197] (c) A hardware circuit and / or a processor, such as a microprocessor or a part of a microprocessor, that requires software (e.g., firmware) to operate, but the software may be absent when not needed for operation.

[0198] The definition of the circuitry applies to all uses of the term in this application, including in any claims. As another example, as used in this application, the term "circuitry" also encompasses an implementation of only a hardware circuit or a processor (or processors) or a part of a hardware circuit or a processor and its (or their) accompanying software and / or firmware. The term "circuitry" also encompasses, for example, an integrated device.

[0199] The foregoing description has provided a complete and informative description of some embodiments by way of example and not limitation. However, various modifications and adaptations may become apparent to those skilled in the relevant art when read in conjunction with the accompanying drawings and the appended claims. However, all such and similar modifications of the teachings will still fall within the scope defined by the appended claims.

Claims

1. A communication method, comprising: receiving, at a machine learning model in a second stage, a first location of a communication device, the first location being determined in a first stage, wherein if a first positioning accuracy of the first location from the first stage is lower than a second positioning accuracy of a positioning service quality, the first location from the first stage is received at the machine learning model in the second stage, wherein the machine learning model in the second stage includes an autoencoder, and the first location from the first stage is received at a latent layer of the autoencoder; if the first positioning accuracy is lower than the second positioning accuracy, receiving the first positioning accuracy from the first stage in the second stage; and in the second stage, determining a second location of the communication device by using the machine learning model based at least on the first location from the first stage and based on the received first positioning accuracy.

2. The method according to claim 1, wherein the first positioning accuracy from the first stage is received at the latent layer of the autoencoder.

3. The method according to claim 1, further comprising: offline training the autoencoder by: calculating a first loss at an output of the autoencoder; calculating a second loss at the latent layer of the autoencoder by using at least the first location of the communication device and a second location of the communication device; and training the autoencoder by using the first loss and the second loss.

4. The method according to claim 3, wherein the second loss includes a sum of a first term and a second term, and the second term enables the second location to be within a space around the first location.

5. The method according to claim 4, wherein the second term is based on a maximum absolute distance between the first location and the second location.

6. A device for communication, comprising: at least one processor; and at least one memory including computer program code, wherein the at least one memory and the computer program code are configured to, together with the at least one processor, cause the device to at least: receive, at a machine learning model in a second stage, a first location of a communication device, the first location being determined in a first stage, wherein if a first positioning accuracy of the first location from the first stage is lower than a second positioning accuracy of a positioning service quality, the first location from the first stage is received at the machine learning model in the second stage, wherein the machine learning model in the second stage includes an autoencoder, and the first location from the first stage is received at a latent layer of the autoencoder; if the first positioning accuracy is lower than the second positioning accuracy, receive the first positioning accuracy from the first stage in the second stage; and In the second stage, determine a second location of the communication device by using the machine learning model, based at least on the first location from the first stage and on the received first positioning accuracy.

7. The apparatus according to claim 6, wherein the first positioning accuracy from the first stage is received in the latent layer of the autoencoder.

8. The apparatus according to claim 6, wherein the at least one memory and the computer program code are further configured to, together with the at least one processor, cause the apparatus to at least: train the autoencoder offline by: calculating a first loss at an output of the autoencoder; calculating a second loss in the latent layer of the autoencoder by using at least the first location of the communication device and the second location of the communication device; and training the autoencoder by using the first loss and the second loss.

9. The apparatus according to claim 8, wherein the second loss comprises a sum of a first term and a second term, the second term causing the second location to be within a space around the first location.

10. The apparatus according to claim 9, wherein the second term is based on a maximum absolute distance between the first location and the second location.

11. A non-transitory computer-readable medium, comprising: program instructions stored thereon for performing the method according to claim 1.

Images