Method and apparatus for positioning, and method and apparatus for training positioning model

Abstract

The present application provides a method and apparatus for positioning, and a method and apparatus for training a positioning model. The method for positioning comprises: acquiring first data, the first data being determined on the basis of CSI data corresponding to a terminal device; inputting the first data into an attention module in a positioning model to obtain first weighted data; and inputting the first weighted data into an output module in the positioning model to obtain location information of the terminal device.

Description

Method and apparatus for positioning, method and apparatus for training positioning model TECHNICAL FIELD

[0001] The present application relates to the technical field of communication, and more particularly, to a method for positioning, a method for training a positioning model, an apparatus for positioning, and an apparatus for training a positioning model. BACKGROUND

[0002] In a communication system, location-based services are increasingly popular. In the related art, a positioning model can be used to position a terminal device. However, the positioning accuracy of the method for positioning based on the positioning model in the related art needs to be further improved.

[0003] SUMMARY

[0004] The present application provides a method for positioning, a method for training a positioning model, an apparatus for positioning, and an apparatus for training a positioning model. Each aspect of the present application is described below.

[0005] In a first aspect, a method for positioning is provided, comprising: obtaining first data, the first data being determined based on CSI data corresponding to a terminal device; inputting the first data into an attention module in a positioning model to obtain first weighted data; inputting the first weighted data into an output module in the positioning model to obtain position information of the terminal device.

[0006] In a second aspect, a method for training a positioning model is provided, the positioning model comprising an attention module and an output module, the method comprising: obtaining first data and a first position label, the first data being determined based on CSI data corresponding to a terminal device, and the first position label corresponding to the first data; inputting the first data into the attention module to obtain first weighted data; inputting the first weighted data into the output module to obtain position information of the terminal device; and training the positioning model according to the position information of the terminal device and the first position label.

[0007] In a third aspect, an apparatus for positioning is provided, comprising: an obtaining module configured to obtain first data, the first data being determined based on CSI data corresponding to a terminal device; an attention module configured to process the first data to obtain first weighted data; and an output module configured to process the first weighted data to obtain position information of the terminal device.

[0008] In a fourth aspect, an apparatus for training a positioning model is provided. The apparatus includes: an obtaining module configured to obtain first data and a first position label, the first data being determined based on CSI data corresponding to a terminal device, and the first position label corresponding to the first data; an attention module configured to process the first data to obtain first weighted data; an output module configured to process the first weighted data to obtain position information of the terminal device; and a training module configured to train the positioning model according to the position information of the terminal device and the first position label.

[0009] In a fifth aspect, an apparatus is provided. The apparatus includes a processor and a memory. The memory is configured to store one or more computer programs. The processor is configured to invoke the computer programs in the memory to cause the terminal device to perform some or all of the steps in the method of the first aspect and / or the second aspect.

[0010] In a sixth aspect, an apparatus is provided. The apparatus includes a processor configured to invoke a program from a memory to cause the apparatus to perform some or all of the steps in the method of the first aspect and / or the second aspect.

[0011] In a seventh aspect, a computer readable storage medium is provided. The computer readable storage medium stores a computer program. The computer program causes the apparatus for training a positioning model and / or the apparatus for positioning to perform some or all of the steps in the method of the various aspects described above.

[0012] In an eighth aspect, a computer program product is provided. The computer program product includes a non-transitory computer readable storage medium storing a computer program. The computer program is operable to cause the apparatus for training a positioning model and / or the apparatus for positioning to perform some or all of the steps in the method of the various aspects described above. In some implementations, the computer program product can be a software installation package.

[0013] In a ninth aspect, a computer program is provided. The computer program causes a computer to perform some or all of the steps in the method of the first aspect and / or the second aspect.

[0014] The positioning model provided by the embodiments of the present application includes an attention module, which can extract global correlation features of first data determined based on CSI data corresponding to a terminal device. Therefore, the positioning accuracy when positioning by the positioning model and the first data will be greatly improved. BRIEF DESCRIPTION OF DRAWINGS

[0015] FIG. 1 is a schematic diagram of a wireless communication system to which the embodiments of the present application are applied.

[0016] FIG. 2 is a structural schematic diagram of a positioning model according to an embodiment of the present application.

[0017] FIG. 3 is a structural schematic diagram of a positioning model according to another embodiment of the present application.

[0018] FIG. 4 is a schematic flowchart of a method for training a positioning model according to an embodiment of the present application.

[0019] FIG. 5 is a schematic flowchart of a method for positioning according to an embodiment of the present application.

[0020] FIG. 6 is a structural schematic diagram of a possible implementation of the positioning model in FIG. 3.

[0021] FIG. 7 is a structural schematic diagram of a possible implementation of a second feature extraction module in the positioning model in FIG. 3.

[0022] FIG. 8 is a structural schematic diagram of a possible implementation of an output module in the positioning model in FIG. 3.

[0023] FIG. 9 is a schematic structural diagram of an apparatus for training a positioning model according to an embodiment of the present application.

[0024] FIG. 10 is a schematic structural diagram of an apparatus for positioning according to an embodiment of the present application.

[0025] FIG. 11 is a schematic structural diagram of an apparatus according to an embodiment of the present application. DETAILED DESCRIPTION

[0026] The technical solutions in the present application will be described below with reference to the accompanying drawings.

[0027] Communication system

[0028] FIG. 1 is a wireless communication system 100 to which embodiments of the present application are applied. The wireless communication system 100 can include a network device 110 and a terminal device 120. The network device 110 can be a device that communicates with the terminal device 120. The network device 110 can provide communication coverage for a specific geographic area and can communicate with the terminal device 120 located in the coverage area.

[0029] FIG. 1 exemplarily shows one network device and two terminals. Optionally, the wireless communication system 100 can include multiple network devices and each network device can include other numbers of terminal devices within its coverage, which is not limited in the embodiments of the present application.

[0030] Optionally, the wireless communication system 100 can further include a network controller, a mobile management entity, and other network entities, which are not limited in the embodiments of the present application.

[0031] It should be understood that the technical solutions of the embodiments of the present application can be applied to various communication systems, for example: a 5th generation (5G) system or new radio (NR), a long term evolution (LTE) system, an LTE frequency division duplex (FDD) system, an LTE time division duplex (TDD) system, and the like. The technical solutions provided in the present application can also be applied to future communication systems, such as a 6th generation mobile communication system, a satellite communication system, and the like.

[0032] The terminal device in the embodiments of the present application can also be referred to as a user equipment (UE), an access terminal, a user unit, a user station, a mobile station, a mobile station (MS), a mobile terminal (MT), a remote station, a remote terminal, a mobile device, a user terminal, a terminal, a wireless communication device, a user agent, or a user apparatus. The terminal device in the embodiments of the present application can refer to a device that provides voice and / or data connectivity for a user, and can be used to connect people, things, and machines, such as handheld devices with wireless connection functions, vehicle-mounted devices, and the like. The terminal device in the embodiments of the present application can be a mobile phone, a tablet computer (Pad), a notebook computer, a palm computer, a mobile internet device (MID), a wearable device, a virtual reality (VR) device, an augmented reality (AR) device, a wireless terminal in industrial control, a wireless terminal in self driving, a wireless terminal in remote medical surgery, a wireless terminal in smart grid, a wireless terminal in transportation safety, a wireless terminal in smart city, a wireless terminal in smart home, and the like. Optionally, the UE can be used to act as a base station. For example, the UE can act as a scheduling entity that provides sidelink signals between UEs in vehicle-to-everything (V2X) or device to device (D2D), and the like. For example, a cellular phone and a car communicate with each other using sidelink signals. The cellular phone and the smart home device communicate with each other without relaying the communication signals through the base station.

[0033] The network device in the embodiments of the present application can be a device for communicating with a terminal device. The network device can also include an access network device. The access network device can also be referred to as a radio access network device or a base station, etc. The access network device in the embodiments of the present application can refer to a radio access network (RAN) node (or device) that accesses a terminal device to a wireless network. The access network device can broadly cover various names in the following or be replaced by the following names, such as: Node B (NodeB), evolved Node B (eNB), next generation Node B (gNB), relay station, access point, transmitting and receiving point (TRP), transmitting point (TP), master eNB (MeNB), secondary eNB (SeNB), multi-standard radio (MSR) node, home base station, network controller, access node, wireless node, access point (AP), transmission node, transceiver node, baseband unit (BBU), remote radio unit (RRU), active antenna unit (AAU), remote radio head (RRH), central unit (CU), distributed unit (DU), positioning node, etc. The base station can be a macro base station, a micro base station, a relay node, a donor node, or the like, or a combination thereof. The base station can also refer to a communication module, modem, or chip used in the aforementioned device or apparatus. The base station can also be a mobile switching center and a device that performs the function of a base station in D2D, V2X, machine-to-machine (M2M) communication, a network side device in a 6G network, a device that performs the function of a base station in a future communication system, etc. The base station can support networks of the same or different access technologies. The embodiments of the present application do not limit the specific technology and specific device form of the access network device.

[0034] The base station can be fixed or mobile. For example, a helicopter or a drone can be configured to act as a mobile base station, and one or more cells can move according to the location of the mobile base station. In other examples, a helicopter or a drone can be configured to act as a device that communicates with another base station.

[0035] In some deployments, the network device in the embodiments of the present application can refer to a CU or a DU, or the network device includes a CU and a DU. The gNB can also include an AAU.

[0036] The network device and the terminal device can be deployed on land, including indoor or outdoor, handheld or vehicle-mounted; can also be deployed on water surface; can also be deployed on aircraft, balloons and satellites in the air. The scenarios in which the network device and the terminal device are located are not limited in the embodiments of the present application.

[0037] The communication devices involved in the wireless communication system can include not only the access network device and the terminal device, but also the core network device. The core network device can also be a kind of network device.

[0038] The core network device in the embodiments of the present application can include devices for processing and forwarding signaling and data of users. For example, the core network device can include core network access and mobility management function (AMF), session management function (SMF), and user plane gateway, positioning server and other core network devices. Among them, the user plane gateway can be a server with functions of mobility management, routing, forwarding and the like for user plane data, and is generally located on the network side, such as serving gateway (SGW) or packet data network gateway (PGW) or user plane function entity (UPF). The AMF and the SMF can be equivalent to the mobility management entity (MME) in the LTE system. The AMF is mainly responsible for admission, and the SMF is mainly responsible for session management. Of course, other network elements can also be included in the core network, which are not listed here.

[0039] The positioning server has a positioning function, and the positioning server involved in the embodiments of the present application can include a location management function (LMF) or a location management component (LMC), or can be a local location management function (LLMF) located in the network device, which is not limited in the embodiments of the present application. In some embodiments, the positioning server can also be referred to as a positioning management device.

[0040] It should be understood that all or part of the functions of the communication device in the present application can also be implemented by software functions running on hardware, or by virtualized functions instantiated on a platform, such as a cloud platform.

[0041] Positioning system

[0042] In a communication system, location-based services (such as navigation positioning, target positioning) are attracting widespread attention, and in view of this, positioning systems are becoming increasingly common. With the development of positioning technology, the positioning system has increasingly high requirements for positioning accuracy, especially for indoor positioning systems.

[0043] In the related art, there is a common positioning system that performs positioning by detectable wireless signals in combination with a positioning model. The detectable wireless signals are, for example, communication signals, ultra-wide band (UWB) signals, wireless fidelity (Wi-Fi) signals, Bluetooth signals, etc.

[0044] The positioning model can also be referred to as a learning network, a deep network, a network, a neural network, or a neural network model. The positioning model can be understood as a model for positioning based on machine learning. There are mainly two methods for positioning based on machine learning: one is a positioning method based on traditional machine learning, and the other is a positioning method based on deep learning.

[0045] The general process of the positioning method based on traditional machine learning is: artificial feature extraction → input of support vector machine (SVM), k-nearest neighbors (KNN) algorithm, etc. → output of position estimation. However, the method based on traditional machine learning needs to perform artificial feature extraction, and for high-dimensional input, it is difficult for humans to efficiently extract reliable features, and the method cannot automatically generalize to new task scenarios using different radio frequency signals, so the positioning accuracy of the method is limited.

[0046] The positioning method based on deep learning can automatically perform feature extraction based on deep learning. Compared with the method based on traditional machine learning, the positioning method based on deep learning can efficiently extract reliable features from input data and is more suitable for processing high-dimensional data.

[0047] In view of the foregoing, the detectable wireless signal contains rich environmental information. For example, the detectable wireless signal can generally include channel state information (CSI) data, and the CSI data generally includes data related to the location of the terminal device. For example, the CSI signal in a multiple-input multiple-output (MIMO) communication system contains spatial information of the network device and the terminal device, that is, the CSI signal contains potential information of the radio frequency environment. In particular, the CSI of a 5G MIMO-orthogonal frequency division multiplexing (MIMO-OFDM) system can provide phase and amplitude information of multiple channel subcarriers, describe the fine-grained characteristics of the channel, and better describe the propagation path of the signal. Therefore, the location information of the terminal device can be obtained by combining the CSI data and the deep learning model.

[0048] In a multiple-input multiple-output communication system, the CSI data is measured by a corresponding reference signal. Suppose a scenario is given, where a base station (BS) has N r antennas, a terminal device (UE) has N t antennas, and various obstacles exist around the base station and the user. Then the discrete Fourier transformation (DFT) form of the received signal can be written as formula (1).

[0049] where H ij [k] represents the channel frequency response between the i-th UE antenna and the j-th base station antenna on the k-th subcarrier. W i [k] represents an additive complex Gaussian noise with a mean of zero and a variance of X j [k] is the signal transmitted by the j-th base station antenna. Y i [k] is the DFT form of the signal received by the i-th UE antenna. It can be seen that the CSI data can be regarded as a three-order tensor That is, the data dimension of the CSI data is at least 3, that is, the CSI data is multi-dimensional data. In some embodiments, the CSI data can also be referred to as CSI or CSI signal.

[0050] Considering that the CSI data is highly structured multi-dimensional data, in order to perform deep learning on the CSI data to obtain the location information of the terminal device, one possible implementation is to use a positioning model based on a convolution neural network (CNN) to learn the CSI data. As shown in FIG. 2, FIG. 2 is a structural schematic diagram of one possible CNN-based positioning model provided by an embodiment of the present application.

[0051] As shown in FIG. 2, the positioning model 200 can include a CNN module 210 and an output module 220.

[0052] The CNN module 210 is constructed by using dense blocks 211 and average pooling layers (AvgPool) 212. Each dense block 211 can include a convolution layer 213, a concatenation layer (concat) 214 and a batch normalization layer (BN) 215. The convolution layer 213 can be, for example, a 2d convolution (Conv2D). The size of the data feature map in each convolution layer 213 remains the same, the concatenation layer 214 can be located between the convolution layers 213, the input data can be jump-propagated based on the convolution layers 213 and the concatenation layers 214, and the batch normalization layer 215 at the tail is used to reduce overfitting. The average pooling layer 212 can be located between the dense blocks 211 and can be used to reduce the size of the feature map, thereby reducing the dimension of the data.

[0053] It should be noted that the dense block 211 shown in FIG. 2 can include 4 convolution layers 213, 2 concatenation layers 214 and one batch normalization layer 215 at the tail. However, the number of layers in the actual dense block 211 can be set as needed.

[0054] The output module 220 can include a flattening layer 221 and a fully connected layer (FCL) 222. The flattening layer 221 is used to flatten the output of the CNN module 210 into a vector, and the fully connected layer 222 is used to map the vector output by the flattening layer 221 to the location information of the terminal device. That is, the fully connected layer 222 is used to output the location information of the terminal device according to the features output by the flattening layer 221.

[0055] The CNN-based positioning model extracts features and information by using trainable filters to convolve multi-dimensional data. It is very efficient in extracting structured features from complex data. However, CNN is mainly used to capture the correlation of adjacent data in multi-dimensional data, that is, CNN is mainly used to capture the local correlation in multi-dimensional data.

[0056] For the convenience of understanding, taking multi-dimensional data as image data as an example, in the field of computer vision, it is generally believed that the spatial relationship of an image is a local pixel relationship, and the correlation of pixels far away from each other is weak, so there is local correlation between adjacent pixels in the image. CNN is good at capturing local correlation in such multi-dimensional data. In view of this, CNN assumes that the features in the image have translational invariance at different positions. This means that when the CNN is processed, the target in the image can be successfully identified regardless of whether it is translated, rotated, scaled, or under different lighting conditions and angles of view.

[0057] CSI data is also multi-dimensional data, but it is significantly different from image data. For example, CSI data is constrained by electromagnetic space, and has not only local correlation but also global correlation. This global correlation can cause great impact on positioning results when rotating or translating the CSI data. However, CNN is difficult to capture global features, so it will ignore the above impact. This results in that when using the above CNN-based positioning model for positioning, the positioning accuracy is difficult to guarantee.

[0058] To solve the above problems, the present application provides a new positioning model, which can include an attention module. When using the positioning model in combination with first data for positioning, the first weighted data can be obtained through the attention module, and the position information of the terminal device can be obtained based on the first weighted data. The first data is determined based on the CSI data corresponding to the terminal device.

[0059] Since the first weighted data contains features representing the global correlation of the CSI data, the positioning model provided in the embodiments of the present application is more suitable for the data characteristics of the CSI data. Therefore, the positioning model provided in the embodiments of the present application can extract features more related to the position of the terminal device based on the attention module, and the features of the multi-dimensional data extracted are more reliable, and the positioning accuracy of the positioning model is further improved, thereby improving the positioning accuracy of the terminal device. In particular, when the positioning model of the present application is used for positioning in an indoor environment, its positioning accuracy is significantly further improved, so the positioning model of the present application is particularly suitable for positioning in a complex multipath environment.

[0060] The positioning model with an attention module provided in the embodiments of the present application will be described schematically in combination with FIG. 3 and FIG. 6.

[0061] As shown in FIG. 3, the positioning model 300 can include an attention module 310 and an output module 320.

[0062] The attention module 310 is configured to process the first data input into the attention module 310 to output first weighted data.

[0063] The first data can be determined based on CSI data corresponding to the terminal device. The CSI data corresponding to the terminal device can be CSI data collected by the terminal device. For example, the CSI data corresponding to the terminal device is obtained by measuring a reference signal by the terminal device. In some embodiments, the CSI data corresponding to the terminal device can be referred to as original CSI data.

[0064] As an implementation manner, the first data can be original CSI data. The original CSI data can be, for example, the original CSI data described above.

[0065] As another implementation manner, the first data can be data obtained by pre-processing the original CSI data, that is, the first data is pre-processed CSI data. The pre-processing can be, for example, one or more of the following: de-noising processing, data integration processing, data compression processing, data expansion processing, smoothing data processing, spatial domain filtering processing, and frequency domain filtering processing.

[0066] As an example, the first data can be data obtained by integrating the original CSI data. For example, the first data can be data obtained by re-integrating the dimensions of the original data. For example, the original CSI data is For example, the first data is wherein, N is an integrated N r and N t The data obtained after N r and N t represent the number of receiving antennas and the number of transmitting antennas, that is, N r and N t represent spatial dimensions, and therefore, N r and N t can be integrated into one dimension. 2 is data obtained by re-setting the first dimension of the first data. The first dimension can also be referred to as a channel dimension. Since the data in the first data includes real and imaginary parts, the number 2 representing the type of real and imaginary parts of the data can be taken as the first dimension. N and K are spatial dimensions of the first data, wherein N is a second dimension of the first data, and K is a third dimension of the first data.

[0067] As another implementation manner, the first data can be data obtained by pre-processing the original CSI data and then extracting features from the pre-processed CSI data. For example, the pre-processed CSI data is For example, the first data can be the data described below wherein, M is greater than 2.

[0068] The type of the reference signal is not specifically limited in the embodiments of the present application, and the reference signal can be any communication signal of a wireless communication system. For example, the reference signal can include, but is not limited to, a 5G signal, a cellular signal, a WiFi signal, a Bluetooth signal, and a future communication signal (for example, a 6G signal).

[0069] The first weighted data can be understood as data obtained by weighting the first data using attention weights. The attention weights are obtained by the attention module 310 processing the first data based on an attention mechanism (or attention algorithm).

[0070] The attention weights can be understood as weights obtained by scoring each data in the first data from a global correlation level. The data in the first data corresponding to a large attention weight can be a feature more relevant to the position of the terminal device.

[0071] The attention module is not specifically limited in the embodiments of the present application, and any network based on an attention mechanism for data processing can be understood as an attention module. For example, the attention module can include one or more of the following modules: a channel attention module and a spatial attention module.

[0072] The output module 320 is configured to process the first weighted data output by the attention module 310 to output the position information of the terminal device.

[0073] The position information of the terminal device can be understood as information that can reflect the position of the terminal device. In some embodiments, the position information of the terminal device can be the position coordinates of the terminal device. In other embodiments, the position information of the terminal device can be a positioning parameter used to calculate the position coordinates of the terminal device, for example, the positioning parameter can be a distance or an angle.

[0074] The output module 320 is not specifically limited in the embodiments of the present application, and the output module can be any neural network that can map multidimensional data to the position information of the terminal device. For example, the output module can be any one or more of the following neural networks: a CNN network and a network including a plurality of residual blocks. In some embodiments, the output module 320 can further include a flattening layer and / or a fully connected layer.

[0075] The positioning model including the attention module provided in the embodiments of the present application can simultaneously extract features of global correlation and features of local correlation according to the characteristics of the CSI data, effectively extracting features more relevant to the position information of the terminal device from the CSI data, and thus the positioning accuracy of the positioning model is effectively improved. In particular, when the positioning model is used in a complex multipath environment, the positioning accuracy of the positioning model in the present application can be significantly better than that of the positioning model in the related art.

[0076] As described above, the embodiments of the present application do not make specific limitations on the attention module, and any network based on the attention mechanism for data processing can be understood as the attention module.

[0077] As an implementation manner, the attention module 310 can include a channel attention module. The channel attention module is configured to process the first data to obtain channel attention weights in the first data. The channel attention weights are configured to weight the first data to obtain first weighted data.

[0078] The embodiments of the present application do not make specific limitations on the processing algorithm of the channel attention module. As an implementation manner, the channel attention module can obtain the channel attention weights based on one or more of the following algorithms: Discrete Cosine Transform (DCT), a fully connected layer, and a Sigmoid activation function.

[0079] In some embodiments, the positioning model can further include a first feature extraction module. The first data can be data obtained by preprocessing the CSI data corresponding to the terminal device by using the first feature extraction module. The first feature extraction module is configured to extract features of the CSI data corresponding to the terminal device to expand the channel dimension of the CSI data. That is, the channel dimension of the first data is greater than the channel dimension of the CSI data corresponding to the terminal device. By expanding the channel dimension of the CSI data, the subsequent fine-grained calculation of the weight of the channel attention module can be facilitated.

[0080] The embodiments of the present application do not make specific limitations on the processing algorithm of the first feature extraction module. As an implementation manner, the first feature extraction module can process the CSI data corresponding to the terminal device based on a CNN network or a DenseNet to obtain the first data.

[0081] As another implementation manner, the attention module 310 can include a spatial attention module. The spatial attention module is configured to process the second data to obtain spatial attention weights in the second data. The second data is data determined based on the first data. For example, the second data is the first data. For another example, the second data is data obtained by processing the first data. The spatial attention weights are configured to weight the second data to obtain the first weighted data.

[0082] In some embodiments, the positioning model can further include a second feature extraction module. The second data can be data obtained by processing the first data using the second feature extraction module. The second feature extraction module is configured to perform spatial dimension processing on the first data to reduce the spatial dimension of the first data. That is, the spatial dimension of the second dimension is smaller than the spatial dimension of the first data. By compressing the spatial dimension of the first data, the computational complexity of the subsequent spatial attention module can be facilitated, thereby improving the computational efficiency of the positioning model.

[0083] As another implementation manner, as shown in FIG. 6, the attention module can include a channel attention module 620 and a spatial attention module 640 at the same time. The channel attention module 620 is configured to obtain channel attention weights in the first data by processing the first data. The channel attention weights are used to weight the first data to obtain third data. The spatial attention module 640 is configured to obtain spatial attention weights in the second data by processing the second data. The second data is data determined based on the third data. For example, the second data is the third data. For another example, the second data is data obtained by processing the third data. The spatial attention weights are used to weight the second data to obtain first weighted data. By simultaneously setting the channel attention module and the spatial attention module, the positioning model can be guided to pay attention to more important parts of the features, reduce the negative impact of multipath and noise on performance, and thereby improve the robustness and accuracy of the position estimation of the terminal device.

[0084] In some embodiments, the positioning model can further include a second feature extraction module. The second data can be data obtained by processing the third data using the second feature extraction module. The second feature extraction module is configured to perform spatial dimension processing on the third data to reduce the spatial dimension of the third data. That is, the spatial dimension of the second dimension is smaller than the spatial dimension of the third data. By compressing the spatial dimension of the third data, the computational complexity of the subsequent spatial attention module can be facilitated, thereby improving the computational efficiency of the positioning model.

[0085] The processing algorithm of the spatial attention module is not specifically limited in the embodiments of the present application. As an implementation manner, the spatial attention module can obtain the spatial attention weights based on one or more of the following algorithms: convolution function and scale transformation.

[0086] The processing algorithm of the second feature extraction module is not specifically limited in the embodiments of the present application. As an implementation manner, the second feature extraction module can obtain the second data based on one or more of the following algorithms: convolution function and pooling function.

[0087] To implement positioning using the positioning model, generally includes a training phase and a prediction phase. The training phase needs to train the positioning model according to a pre-obtained collection data set to obtain a learning model; the prediction phase needs to use the trained positioning model to estimate the position information of the terminal device according to the collected CSI data, and finally output the position information of the terminal device.

[0088] Therefore, the embodiments of the present application also provide a method for training a positioning model and a method for positioning. The method for positioning is a method for positioning based on a positioning model. The positioning model in the method embodiments is the same as the positioning model comprising the attention module described above, and therefore, the same parts will not be described again, and reference can be made to the foregoing.

[0089] The method for training a positioning model provided by the embodiments of the present application will be introduced below in combination with FIG. 4. That is, FIG. 4 is a flowchart of the training phase. The method shown in FIG. 4 can be executed by an electronic device. The electronic device may, for example, be the communication device described above. For example, the method shown in FIG. 4 can be executed by one or more of the positioning server, the network device and the terminal device. As shown in FIG. 4, the method can include steps S410-S440.

[0090] Step S410, obtaining first data and a first position label.

[0091] The first position label corresponds to the first data. The first position label can be understood as the real position information of the terminal device corresponding to the first data. As described above, the first data is determined based on the CSI data corresponding to the terminal device. The first position label and the first data together form the labeled data (or training data) for training the positioning model. The correspondence between the first position label and the first data may, for example, be obtained by experimental testing in an indoor environment.

[0092] The embodiments of the present application do not make specific limitations on the manner of obtaining the first data.

[0093] As an implementation manner, the first data can be the original CSI data obtained by the terminal device through measurement of the corresponding reference signal, that is, the CSI data corresponding to the terminal device.

[0094] As another implementation manner, the first data can be data obtained from the output of the pre-processing module and / or the first feature extraction module. The pre-processing module and / or the first feature extraction module can process the original CSI data to output the first data.

[0095] Step S420, inputting the first data into the attention module in the positioning model to obtain first weighted data.

[0096] The attention module in the positioning model is input with the first data to obtain first weighted data. It can be understood that the input of the attention module is the first data, and the output is the first weighted data.

[0097] In step S430, the output module in the positioning model is input with the first weighted data to obtain the position information of the terminal device.

[0098] The output module in the positioning model is input with the first weighted data to obtain the position information of the terminal device. It can be understood that the input of the output module is the first weighted data, and the output is the position information of the terminal device.

[0099] In step S440, the positioning model is trained according to the position information of the terminal device and the first position label.

[0100] The positioning model is trained according to the position information of the terminal device and the first position label. It can be understood that the parameters in the positioning model are adjusted according to the position information of the terminal device and the first position label. The parameters in the positioning model can be the parameters of each module in the positioning model, so that the accuracy of the positioning model is higher.

[0101] For example, the loss result can be calculated according to the position information of the terminal device and the first position label, and it is determined whether to adjust the parameters of the positioning model according to whether the loss result meets the requirement. The method of adjusting the parameters of the positioning model can be, for example, updating the parameters of the positioning model in a back propagation manner based on the loss result, so that the model parameters of the positioning model converge in the direction of reducing the loss result.

[0102] The positioning model trained by the method for training a positioning model provided in the application can extract and learn the global correlation features of the first data determined based on the CSI data corresponding to the terminal device, that is, the features more related to the position of the terminal device. Therefore, the positioning model is more accurate.

[0103] In some embodiments, the attention module includes a channel attention module, and step S420 includes: processing the first data by using the channel attention module to obtain the weight of the channel feature in the first data; and weighting the first data by using the weight of the channel feature to obtain the first weighted data. That is, when the attention module only includes the channel attention module, the input of the channel attention module is the first data, and the output of the channel attention module is the first weighted data.

[0104] In some embodiments, the attention module includes a spatial attention module, and step S420 includes: processing the second data by using the spatial attention module to obtain a weight of a spatial feature in the second data, the second data being determined based on the first data; and weighting the second data by using the weight of the spatial feature to obtain the first weighted data. That is, when the attention module only includes the spatial attention module, the second data is the input of the spatial attention module, and the first weighted data is the output of the spatial attention module. The second data is determined based on the first data. For example, the second data is the first data. For another example, the second data is data obtained by transforming the spatial dimension of the first data. As an example, the second data can be data obtained by compressing the spatial dimension of the first data.

[0105] In some embodiments, the attention module includes a channel attention module and a spatial attention module, and step S420 includes: processing the first data by using the channel attention module to obtain a weight of a channel feature in the first data; weighting the first data by using the weight of the channel feature to obtain third data; processing the second data by using the spatial attention module to obtain a weight of a spatial feature in the second data, the second data being determined based on the third data; and weighting the second data by using the weight of the spatial feature to obtain the first weighted data. That is, when the attention module only includes the channel attention module and the spatial attention module, the first data is the input of the channel attention module, and the third data is the output of the channel attention module. The second data is the input of the spatial attention module, and the first weighted data is the output of the spatial attention module. The second data is determined based on the third data. For example, the second data is the third data. For another example, the second data is data obtained by transforming the spatial dimension of the third data. As an example, the second data can be data obtained by compressing the spatial dimension of the third data.

[0106] In some embodiments, the data to be input into the channel attention module can be preprocessed. As an implementation manner, the positioning model includes a first feature extraction module, and the method of training the positioning model further includes: processing the CSI data by using the first feature extraction module to obtain the first data, the channel dimension of the first data being greater than the channel dimension of the CSI data. The first data is the data to be input into the channel attention module, and the CSI data is the output data of the previous module of the channel attention module, i.e., the data to be input into the channel attention module. For example, the previous module of the channel attention module is the acquisition module, and the CSI data is the original CSI data. The first feature extraction module can be helpful for the subsequent fine-grained calculation of the channel attention module.

[0107] In some embodiments, the data to be input into the spatial attention module can be preprocessed. As an implementation manner, the positioning model comprises a second feature extraction module, and the method of training the positioning model further comprises: processing the first data or the third data by using the second feature extraction module to obtain second data, the spatial dimension of the second data being smaller than the spatial dimension of the first data or the third data. The second data is the data to be input into the spatial attention module, and the CSI data is the output data of the previous module of the spatial attention module, i.e., the data to be input into the spatial attention module. For example, the previous module of the spatial attention module is the acquisition module, and the second data is the first data (e.g., the original CSI data). For another example, the previous module of the spatial attention module is the channel attention module, and the second data is the third data. The second feature extraction module can help reduce the calculation amount of the subsequent spatial attention module, thereby improving the positioning efficiency.

[0108] FIG. 5 is a schematic flowchart of a method for positioning provided by an embodiment of the present application. That is, FIG. 5 is a flowchart of the prediction stage. The method shown in FIG. 5 can be performed by an electronic device. The electronic device can be the same as the electronic device 5 in the training stage. The method shown in FIG. 5 can comprise steps S510-S530.

[0109] In step S510, first data is obtained.

[0110] As described above, the first data is determined based on the CSI data corresponding to the terminal device.

[0111] The present embodiment does not make specific limitations on the manner of obtaining the first data.

[0112] As an implementation manner, the first data can be original CSI data obtained by the terminal device through measurement of a corresponding reference signal, the original CSI data being the CSI data corresponding to the terminal device.

[0113] As another implementation manner, the first data can be data obtained from the output of a preprocessing module and / or a first feature extraction module, the preprocessing module and / or the first feature extraction module can process the original CSI data to output the first data.

[0114] In step S520, the first data is input into an attention module in the positioning model to obtain first weighted data.

[0115] Inputting the first data into the attention module in the positioning model to obtain the first weighted data can be understood as that the input of the attention module is the first data, and the output is the first weighted data.

[0116] In step S530, the first weighted data is input into an output module in the positioning model to obtain the position information of the terminal device.

[0117] The first weighted data is input into an output module in the positioning model to obtain the position information of the terminal device. It can be understood that the input of the output module is the first weighted data, and the output is the position information of the terminal device.

[0118] The positioning method provided in the embodiments of the present application can be used for positioning by a positioning model including an attention module. The positioning model can extract global correlation features of the first data determined based on the CSI data corresponding to the terminal device, that is, features more relevant to the position of the terminal device are extracted and learned. Therefore, the positioning accuracy when positioning by the positioning model and the first data will be greatly improved.

[0119] In some embodiments, the attention module includes a channel attention module, and step S520 includes: processing the first data by using the channel attention module to obtain weights of channel features in the first data; and weighting the first data by using the weights of the channel features to obtain the first weighted data. That is, when the attention module only includes the channel attention module, the first data is the input of the channel attention module, and the first weighted data is the output of the channel attention module.

[0120] In some embodiments, the attention module includes a spatial attention module, and step S520 includes: processing the second data by using the spatial attention module to obtain weights of spatial features in the second data, the second data being determined based on the first data; and weighting the second data by using the weights of the spatial features to obtain the first weighted data. That is, when the attention module only includes the spatial attention module, the second data is the input of the spatial attention module, and the first weighted data is the output of the spatial attention module. The second data is determined based on the first data. For example, the second data is the first data. For another example, the second data is data obtained by transforming the spatial dimension of the first data. As an example, the second data can be data obtained by compressing the spatial dimension of the first data.

[0121] In some embodiments, the attention module includes a channel attention module and a spatial attention module, and step S520 includes: processing the first data by using the channel attention module to obtain a weight of a channel feature in the first data; weighting the first data by using the weight of the channel feature to obtain third data; processing the second data by using the spatial attention module to obtain a weight of a spatial feature in the second data, the second data being determined based on the third data; and weighting the second data by using the weight of the spatial feature to obtain first weighted data. That is, when the attention module only includes the channel attention module and the spatial attention module, the first data is the input of the channel attention module, and the third data is the output of the channel attention module. The second data is the input of the spatial attention module, and the first weighted data is the output of the spatial attention module. The second data is determined based on the third data. For example, the second data is the third data. For another example, the second data is data obtained by transforming the spatial dimension of the third data. As an example, the second data can be data obtained by compressing the spatial dimension of the third data.

[0122] In some embodiments, the data to be input into the channel attention module can be preprocessed. As an implementation manner, the positioning model includes a first feature extraction module, and the method of training the positioning model further includes: processing the CSI data by using the first feature extraction module to obtain the first data, the channel dimension of the first data being greater than the channel dimension of the CSI data. The first data is the data to be input into the channel attention module, and the CSI data is the output data of the previous module of the channel attention module, that is, the data to be input into the channel attention module. For example, the previous module of the channel attention module is the acquisition module, and the CSI data is the original CSI data. The first feature extraction module can help the subsequent channel attention module to perform fine-grained calculation.

[0123] In some embodiments, the data to be input into the spatial attention module can be preprocessed. As an implementation manner, the positioning model includes a second feature extraction module, and the method of training the positioning model further includes: processing the first data or the third data by using the second feature extraction module to obtain the second data, the spatial dimension of the second data being less than the spatial dimension of the first data or the third data. The second data is the data to be input into the spatial attention module, and the CSI data is the output data of the previous module of the spatial attention module, that is, the data to be input into the spatial attention module. For example, the previous module of the spatial attention module is the acquisition module, and the second data is the first data (for example, the original CSI data). For another example, the previous module of the spatial attention module is the channel attention module, and the second data is the third data. The second feature extraction module can help to reduce the calculation amount of the subsequent spatial attention module, thereby improving the positioning efficiency.

[0124] The embodiments of the positioning model in the present application will be described in more detail below in combination with specific examples of FIG. 6-FIG. 8. It should be noted that the examples of FIG. 6-FIG. 8 are only to help those skilled in the art to understand the embodiments of the present application, and are not intended to limit the embodiments of the present application to the specific values or specific scenarios illustrated. Those skilled in the art can obviously make various equivalent modifications or changes according to the examples of FIG. 6-FIG. 8 given, and such modifications or changes also fall within the scope of the embodiments of the present application.

[0125] As shown in FIG. 6, the positioning model includes a first feature extraction module 610, a channel attention module 620, a second feature extraction module 630, a spatial attention module 640, and an output module 650.

[0126] The input of the first feature extraction module 610 is data a. The data a can be data obtained after pre-processing the original CSI data or the original CSI data. As an example, the original CSI data is The data a is wherein N can be determined according to N r and N t .

[0127] The first feature extraction module 610 can process the input data a to obtain first data. That is, the first data is the output data of the first feature extraction module 610. In some embodiments, the first feature extraction module also becomes an input part or an input block.

[0128] The embodiments of the present application do not make specific limitations on the first feature extraction module 610, for example, the first feature extraction module 610 is a DenseNet, and the first data is the output data of the DenseNet after the data a is input into the DenseNet.

[0129] The first data can be represented as which can be obtained by the following formula 2.

[0130] wherein, M is greater than 2. M is the number of new dimensions after feature propagation.

[0131] The first feature extraction module 610 can perform fine-grained processing on the data a and gradually increase the horizontal slice dimension number (i.e., the channel feature dimension or the channel dimension or the first dimension) of the input tensor, thereby promoting the propagation of CSI features. In addition, the first feature extraction module 610 can provide better weighting results for the first dimension of the data a. The expansion of the first dimension is conducive to the weighting of subsequent network channels and the attention between different network channels, and is conducive to extracting location-related potential features or information.

[0132] The channel attention module 620 can also be referred to as a channel attention block. The input of the channel attention module 620 is connected with the output of the first feature extraction module 610, that is, the first data is the input data of the channel attention module 620. The channel attention module 620 is mainly used to quantify and weigh the importance between different channel features. The processing method of the channel attention module 620 is as follows.

[0133] First, the first data is divided into L cubic bodies of equal size along the channel dimension, and then wherein

[0134] Then, each component is assigned a corresponding DCT frequency component. The result of the two-dimensional DCT can be used as a compressed representation of the channel attention of each component. The result q of the two-dimensional DCT (l) can be determined based on formula 3.

[0135] wherein, is the two-dimensional index of the frequency component, corresponding to

[0136] The entire frequency domain vector q can be determined based on formula 4.

[0137] wherein, q ∈ R M .

[0138] The entire channel attention vector c Att can be determined based on formula 5.

[0139] wherein represents a fully connected layer, and ρ(·) represents a Sigmoid activation function.

[0140] The channel attention vector is used to weight the first feature, that is, the weight of the channel feature in the first data described above. By using the channel attention vector to weight the first feature, the effectiveness of feature expression and capture can be improved.

[0141] The data after weighting the first data using the channel attention vector (i.e., the third data described above) can be determined based on formula 6.

[0142] wherein is the i-th element of the channel attention vector c Att The channel feature corresponding to the feature with more information has a higher weight.

[0143] The second feature extraction module 630 can be a deep feature extraction block (DFEB). The deep feature extraction block can be, for example, a CNN deep network or other types of deep neural learning network. The input of the second feature extraction module 630 is connected with the output of the channel attention module 620, that is, the third data is the input of the second feature extraction module 630. The main purpose of the second feature extraction module 630 is to integrate the shallow features to obtain deep features and reduce the spatial dimension.

[0144] The data output by the second feature extraction module 630 is second data, which can be expressed as The spatial dimension can be understood as reducing the value of the second dimension N and the third dimension K in that is, less than N, less than K.

[0145] The second data can be determined based on formula 7.

[0146] As an example, as shown in FIG. 7, the second feature extraction module 630 can include stacking N basic blocks 631 to obtain deep features, thereby facilitating the input of the subsequent spatial attention module 640. Wherein, N is greater than or equal to 2. Each basic block can include a residual connection (or input layer) for fusing shallow information while preserving the depth of the network. Deep networks (for example, the deep network in FIG. 7 can include Conv1, Conv2, Conv3 and Conv4) can introduce more nonlinear factors in the network, which means they can simulate more complex functions. This is particularly useful for indoor positioning. The pooling layer (for example, the average pooling layer in FIG. 7) is used to gradually reduce the spatial size of the features.

[0147] The spatial attention module 640 can also be referred to as a spatial attention block. The input of the spatial attention module 640 is connected with the output of the second feature extraction module 630, that is, the second data is the input data of the spatial attention module 640. The spatial attention module 640 is mainly used to understand which pixels in the second data are more important. As is known, the effective receptive field of the CNN follows a Gaussian distribution, which means that the center pixel in the receptive field usually has a greater impact. By introducing a spatial attention mechanism, the neural network can see the entire input, rather than just focusing on a local area. The processing method of the spatial attention module 640 is as follows.

[0148] First, 3 intermediate features are obtained by 1x1 convolution, value

[0149] Then, the data b is obtained by a re-shape operation.

[0150] By simultaneously considering the pixel position and the feature similarity, the attention output ratio is converted into a probability, and thus the data b can be derived. The data b can be determined based on formula 8.

[0151] wherein, is a weight of the spatial feature in the second data. The G can be determined based on formula 9.

[0152] wherein, are the relative position matrices of the height and the width, respectively.

[0153] Then, the data b is reshaped into a tensor with dimensions and the output data of the whole spatial attention module 640 is obtained by a convolution layer. i.e., the first weighted data. The first weighted data can be determined by formula 10.

[0154] The input of the output block 650 is connected with the output of the spatial attention module 640, that is, the first weighted data is the input data of the output block 650. The output block 650 is configured to further integrate deep information to obtain a two-dimensional coordinate representing the position information of the terminal device from the high-dimensional feature data.

[0155] The embodiments of the present application do not make specific limitations on the output block 650. For example, the output block 650 can be a CNN or other neural network that can output a two-dimensional coordinate representing the position information of the terminal device from high-dimensional data.

[0156] As an example, as shown in FIG. 8, the output block 650 can include a residual block 651, a flattening layer 652, and a fully connected layer 653.

[0157] The residual block 651 can be L residual blocks stacked together, i.e., L residual blocks connected in series, and L is greater than or equal to 2. Each residual block can include an input layer, a Conv1, and a Conv2. In addition, each residual block can adopt a skip connection, which provides a convenient way for gradient-based backpropagation, helps to stabilize the training phase and reduce the training pressure.

[0158] The flattening layer 652 is configured to flatten the output data of the residual block 651 into a vector. The flattened vector can be understood as a deep feature.

[0159] The fully connected layer 653 is configured to gradually map the flattened vector from a high-dimensional feature space to the position information of the terminal device. That is, the output of the fully connected layer 653 is the position information of the terminal device finally output by the output part 650.

[0160] If the position information of the terminal device is information directly used to indicate the position coordinates of the terminal device, the fully connected layer 653 will gradually map the high-dimensional feature space to two-dimensional coordinates. If the position information of the terminal device is a positioning parameter used to calculate the position coordinates of the terminal device, for example, the positioning parameter can be a distance or an angle, the fully connected layer 653 will gradually map the high-dimensional feature space to coordinates or indicators corresponding to the positioning parameter.

[0161] The position information Pos of the terminal device output by the fully connected layer 653 can be determined based on formula 11.

[0162] The positioning scheme based on the above positioning model includes a training phase and a prediction phase. The training phase needs to train the preset positioning model according to the pre-obtained collection data set to obtain a stable positioning model; the prediction phase needs to use the trained network model to estimate the position information of the terminal device according to the collected CSI data, and finally output the position information of the terminal device.

[0163] The main purpose of the training phase (pre-training phase) is to construct a database and train a deep neural network, and the main process is as follows.

[0164] First, a measurement system is built, and CSI data required for experimental testing and collection during training and position labels corresponding to the CSI data are collected in an indoor environment, and an annotated data set in the environment is established for training the positioning model.

[0165] Based on the collection data set constructed in the first step, a training set and a verification set are obtained, and then the designed deep learning model is trained. The input of the model is the CSI data of the collected signal, and the output is the estimated position of the user.

[0166] The model is updated in parameters according to the error between the estimated position information of the terminal device by the model and the position label of the CSI data, and the network parameters are iteratively updated until the training converges. The loss function for iteratively updating the network parameters is the root mean square error (RMSE) of the position error.

[0167] The prediction stage (actual use stage) of the system mainly aims to realize accurate estimation of the position information of the terminal device through network parameters and collected data. The main process is: collecting CSI data of other positions except the training set, inputting the model to estimate and calculate the position information of the terminal device, and finally outputting the accurate position of the terminal device.

[0168] The method embodiments of the present application are described in detail above, and the device embodiments of the present application are described in detail below. It should be understood that the description of the method embodiments corresponds to the description of the device embodiments, and therefore, the parts not described in detail can be referred to the foregoing method embodiments.

[0169] FIG. 9 is a schematic structural diagram of a device 900 for training a positioning model according to an embodiment of the present application. The device 900 comprises an acquisition module 910, an attention module 920, an output module 930, and a training module 940.

[0170] The acquisition module 910 is configured to acquire first data and a first position label, wherein the first data is determined based on CSI data corresponding to a terminal device, and the first position label corresponds to the first data.

[0171] The attention module 920 is configured to process the first data to obtain first weighted data.

[0172] The output module 930 is configured to process the first weighted data to obtain position information of the terminal device.

[0173] The training module 940 is configured to train the positioning model according to the position information of the terminal device and the first position label.

[0174] Optionally, the attention module 920 comprises a channel attention module, which is configured to process the first data to obtain a weight of a channel feature in the first data, and weight the first data by using the weight of the channel feature to obtain the first weighted data.

[0175] Optionally, the device comprises a first feature extraction module, which is configured to process the CSI data to obtain the first data, wherein a channel dimension of the first data is greater than a channel dimension of the CSI data.

[0176] Optionally, the attention module further comprises a spatial attention module, which is configured to process the second data to obtain a weight of a spatial feature in the second data, wherein the second data is determined based on the first data or third data, the third data is data obtained by weighting the first data by using the weight of the channel feature, and the spatial attention module is configured to weight the second data by using the weight of the spatial feature to obtain the first weighted data.

[0177] Optionally, the apparatus comprises a second feature extraction module configured to process the first data or the third data to obtain the second data, wherein a spatial dimension of the second data is less than a spatial dimension of the first data or the third data.

[0178] Fig. 10 is an apparatus 1000 for positioning according to an embodiment of the application. As an implementation form, the apparatus 1000 can be a communication device. The apparatus 1000 comprises an obtaining module 1010, an attention module 1020 and an output module 1030.

[0179] The obtaining module 1010 is configured to obtain first data, wherein the first data is determined based on CSI data corresponding to a terminal device.

[0180] The attention module 1020 is configured to process the first data to obtain first weighted data.

[0181] The output module 1030 is configured to process the first weighted data to obtain position information of the terminal device.

[0182] Optionally, the attention module comprises a channel attention module configured to process the first data to obtain a weight of a channel feature in the first data, and to weight the first data using the weight of the channel feature to obtain the first weighted data.

[0183] Optionally, the apparatus further comprises a first feature extraction module configured to process the CSI data to obtain the first data, wherein a channel dimension of the first data is greater than a channel dimension of the CSI data.

[0184] Optionally, the attention module further comprises a spatial attention module configured to process second data to obtain a weight of a spatial feature in the second data, wherein the second data is determined based on the first data or third data, the third data being obtained by weighting the first data using the weight of the channel feature, and to weight the second data using the weight of the spatial feature to obtain the first weighted data.

[0185] Optionally, the apparatus comprises a second feature extraction module configured to process the first data or the third data to obtain the second data, wherein a spatial dimension of the second data is less than a spatial dimension of the first data or the third data.

[0186] FIG. 11 is a schematic structural diagram of an apparatus according to an embodiment of the present application. The apparatus 1100 can be a device for training a positioning model, or a device for positioning. Alternatively, the apparatus 1100 can be a communication device. The dashed lines in FIG. 11 indicate that the unit or module is optional. The apparatus 1100 can be used to implement the method described in the above method embodiments. The apparatus 1100 can be a pin device, a chip, a terminal device, or a network device.

[0187] The apparatus 1100 can include one or more processors 1110. The processor 1110 can support the apparatus 1100 to implement the method described in the above method embodiments. The processor 1110 can be a general purpose processor or a special purpose processor. For example, the processor can be a central processing unit (CPU). Alternatively, the processor can also be other general purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic components, discrete hardware components, etc. The general purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0188] The apparatus 1100 can also include one or more memories 1120. The memory 1120 stores a program, which can be executed by the processor 1110, so that the processor 1110 performs the method described in the above method embodiments. The memory 1120 can be independent of the processor 1110 or integrated in the processor 1110.

[0189] The apparatus 1100 can also include a transceiver 1130. The processor 1110 can communicate with other devices or chips through the transceiver 1130. For example, the processor 1110 can perform data transceiving with other devices or chips through the transceiver 1130.

[0190] The embodiments of the present application also provide a computer readable storage medium for storing a program. The computer readable storage medium can be applied to the apparatus provided by the embodiments of the present application, and the program causes the computer to execute the method in each of the embodiments of the present application.

[0191] The embodiments of the present application also provide a computer program product. The computer program product includes a program. The computer program product can be applied to the apparatus provided by the embodiments of the present application, and the program causes the computer to execute the method in each of the embodiments of the present application.

[0192] The embodiments of the present application further provide a computer program. The computer program can be applied to the apparatus provided by the embodiments of the present application, and the computer program enables a computer to execute the method in the embodiments of the present application.

[0193] It should be understood that the terms "system" and "network" can be used interchangeably in the present application. In addition, the terms used in the present application are only used to explain the specific embodiments of the present application, and are not intended to limit the present application. The terms "first", "second", "third", and "fourth" and the like in the description and claims of the present application and the drawings are used to distinguish different objects, and are not used to describe a particular order. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion.

[0194] In the embodiments of the present application, the "indication" mentioned can be direct indication, or indirect indication, or can be an indication of an associated relationship. For example, A indicates B, which can mean that B can be obtained directly through A; or A indirectly indicates B, for example, A indicates C, and B can be obtained through C; or A and B have an associated relationship.

[0195] In the embodiments of the present application, "B corresponding to A" means that B is associated with A, and B can be determined according to A. However, it should also be understood that determining B according to A does not mean that B is determined only according to A, but B can also be determined according to A and / or other information.

[0196] In the embodiments of the present application, the term "corresponding" can mean that there is a direct or indirect corresponding relationship between the two, or can mean that there is an associated relationship between the two, or can mean an indication and being indicated, configuration and being configured, and the like.

[0197] In the embodiments of the present application, "predefined" or "preconfigured" can be realized by pre-saving corresponding codes, tables or other means for indicating related information in devices (for example, including terminal devices and network devices), and the present application does not limit the specific implementation manner. For example, predefinition can mean definition in a protocol.

[0198] In the embodiments of the present application, the "protocol" can refer to a standard protocol in the communication field, for example, can include an LTE protocol, an NR protocol, and a related protocol applied to a future communication system, and the present application does not limit this.

[0199] In the embodiments of the present application, the term "and / or" is only a description of the associated relationship between the associated objects, which means that there can be three relationships, for example, A and / or B, which can mean that A exists alone, A and B exist together, and B exists alone. In addition, the character " / " in this paper generally means that the front and rear associated objects have an "or" relationship.

[0200] In the embodiments of the present application, the "comprising" can mean directly comprising or indirectly comprising. Alternatively, the "comprising" mentioned in the embodiments of the present application can be replaced by "indicating" or "for determining". For example, A comprising B can be replaced by A indicating B, or A for determining B.

[0201] In various embodiments of the present application, the size of the serial number of the above processes does not mean the order of execution, and the execution order of the processes should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0202] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interface, device or unit, and can be electrical, mechanical or other forms.

[0203] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on a plurality of network units. According to actual needs, part or all of the units can be selected to achieve the purpose of the present embodiment.

[0204] In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit.

[0205] In the above embodiments, all or part of the embodiments can be implemented by software, hardware, firmware or any combination thereof. When implemented by software, all or part of the embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present application are generated. The computer can be a general purpose computer, a special purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer readable storage medium or transmitted from one computer readable storage medium to another computer readable storage medium, for example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center through wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) mode. The computer readable storage medium can be any available medium that can be read by a computer or a data storage device such as a server, data center and the like integrated with one or more available media sets. The available media can be magnetic media (for example, floppy disk, hard disk, magnetic tape), optical media (for example, digital video disc (DVD)) or semiconductor media (for example, solid state disk (SSD)) and the like.

[0206] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical range disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for positioning, characterized by, The method comprises: obtaining first data, wherein the first data is determined based on CSI data corresponding to a terminal device; inputting the first data into an attention module in a positioning model to obtain first weighted data; inputting the first weighted data into an output module in the positioning model to obtain position information of the terminal device.

2. The method of claim 1, wherein, The attention module comprises a channel attention module, and the inputting the first data into the attention module in the positioning model to obtain the first weighted data comprises: processing the first data by using the channel attention module to obtain weights of channel features in the first data; weighting the first data by using the weights of the channel features to obtain the first weighted data.

3. The method of claim 2, wherein, The positioning model comprises a first feature extraction module, and the method further comprises: processing the CSI data by using the first feature extraction module to obtain the first data, wherein a channel dimension of the first data is greater than a channel dimension of the CSI data.

4. The method according to claim 1 or 2, characterized in that, The attention module further comprises a spatial attention module, and the inputting the first data into the attention module in the positioning model to obtain the first weighted data comprises: processing second data by using the spatial attention module to obtain weights of spatial features in the second data, wherein the second data is determined based on the first data or third data, and the third data is data obtained by weighting the first data by using the weights of the channel features; weighting the second data by using the weights of the spatial features to obtain the first weighted data.

5. The method of claim 4, wherein, The positioning model comprises a second feature extraction module, and the method further comprises: processing the first data or the third data by using the second feature extraction module to obtain the second data, wherein a spatial dimension of the second data is less than a spatial dimension of the first data or the third data.

6. A method of training a localization model, the method comprising: The positioning model comprises an attention module and an output module, and the method comprises: obtaining first data and a first position label, wherein the first data is determined based on CSI data corresponding to a terminal device, and the first position label corresponds to the first data; inputting the first data into the attention module to obtain first weighted data; inputting the first weighted data into the output module to obtain position information of the terminal device; training the positioning model according to the position information of the terminal device and the first position label.

7. The method of claim 6, wherein, The attention module comprises a channel attention module, and the inputting the first data into the attention module in the positioning model to obtain the first weighted data comprises: processing the first data by using the channel attention module to obtain weights of channel features in the first data; weighting the first data by using the weights of the channel features to obtain the first weighted data.

8. The method of claim 7, wherein, The positioning model comprises a first feature extraction module, and the method further comprises: processing the CSI data by using the first feature extraction module to obtain the first data, wherein a channel dimension of the first data is greater than a channel dimension of the CSI data.

9. The method according to claim 6 or 7, characterized in that, The attention module further includes a spatial attention module, the attention module in the positioning model for processing the first data, to obtain the first weighted data including: processing the second data by using the spatial attention module to obtain the weight of the spatial feature in the second data, the second data being determined based on the first data or third data, the third data being data obtained by weighting the first data by using the weight of the channel feature; weighting the second data by using the weight of the spatial feature to obtain the first weighted data.

10. The method of claim 9, wherein, The positioning model includes a second feature extraction module, and the method further includes: processing the first data or the third data by using the second feature extraction module to obtain the second data, the spatial dimension of the second data being less than the spatial dimension of the first data or the third data.

11. An apparatus for positioning, characterized by including: an acquisition module, configured to acquire first data, the first data being determined based on CSI data corresponding to a terminal device; an attention module, configured to process the first data to obtain first weighted data; an output module, configured to process the first weighted data to obtain position information of the terminal device.

12. The apparatus of claim 11, wherein, The attention module includes a channel attention module, and the channel attention module is configured to: process the first data to obtain the weight of a channel feature in the first data; weight the first data by using the weight of the channel feature to obtain the first weighted data.

13. The apparatus of claim 12, wherein, The device further includes: a first feature extraction module, configured to process the CSI data to obtain the first data, the channel dimension of the first data being greater than the channel dimension of the CSI data.

14. The apparatus of claim 11 or 12, wherein, The attention module further includes a spatial attention module, and the spatial attention module is configured to: process second data to obtain the weight of a spatial feature in the second data, the second data being determined based on the first data or third data, the third data being data obtained by weighting the first data by using the weight of the channel feature; weight the second data by using the weight of the spatial feature to obtain the first weighted data.

15. The apparatus of claim 14, wherein, The device includes a second feature extraction module, configured to process the first data or the third data to obtain the second data, the spatial dimension of the second data being less than the spatial dimension of the first data or the third data.

16. An apparatus for training a localization model, the apparatus comprising: The device includes: an acquisition module, configured to acquire first data and a first position label, the first data being determined based on CSI data corresponding to a terminal device, the first position label corresponding to the first data; an attention module, configured to process the first data to obtain first weighted data; an output module, configured to process the first weighted data to obtain position information of the terminal device; a training module, configured to train the positioning model according to the position information of the terminal device and the first position label.

17. The apparatus of claim 16, wherein, The attention module includes a channel attention module, and the channel attention module is configured to: process the first data to obtain the weight of a channel feature in the first data; The first data is weighted by using the weight of the channel feature to obtain the first weighted data.

18. The apparatus of claim 17, wherein, The device comprises a first feature extraction module, which is configured to: The CSI data is processed to obtain the first data, and the channel dimension of the first data is greater than the channel dimension of the CSI data.

19. The apparatus of claim 16 or 17, wherein, The attention module further comprises a spatial attention module, which is configured to: The second data is processed to obtain the weight of the spatial feature in the second data, and the second data is determined based on the first data or the third data, and the third data is obtained by weighting the first data by using the weight of the channel feature; The second data is weighted by using the weight of the spatial feature to obtain the first weighted data.

20. The apparatus of claim 19, wherein, The device comprises a second feature extraction module, which is configured to: The first data or the third data is processed to obtain the second data, and the spatial dimension of the second data is less than the spatial dimension of the first data or the third data.

21. An apparatus, comprising: The device comprises a memory and a processor, wherein the memory is configured to store a program, and the processor is configured to invoke the program in the memory to enable the device to perform the method in any one of claims 1-5 or 6-10.

22. An apparatus comprising: The device comprises a processor configured to invoke a program from a memory to enable the device to perform the method in any one of claims 1-5 or 6-10.

23. A computer-readable storage medium, characterized in that, The computer has a program stored thereon, and the program enables the computer to perform the method in any one of claims 1-5 or 6-10.

24. A computer program product, characterised in that, The computer has a program stored thereon, and the program enables the computer to perform the method in any one of claims 1-5 or 6-10.

25. A computer program, characterized in that, The computer program enables the computer to perform the method in any one of claims 1-5 or 6-10.

Images