WO2023071760A1 - 波束域的划分方法及装置、存储介质及电子装置 - Google Patents

波束域的划分方法及装置、存储介质及电子装置 Download PDF

Info

Publication number
WO2023071760A1
WO2023071760A1 PCT/CN2022/124391 CN2022124391W WO2023071760A1 WO 2023071760 A1 WO2023071760 A1 WO 2023071760A1 CN 2022124391 W CN2022124391 W CN 2022124391W WO 2023071760 A1 WO2023071760 A1 WO 2023071760A1
Authority
WO
WIPO (PCT)
Prior art keywords
measurement data
label
beam domain
model
data
Prior art date
Application number
PCT/CN2022/124391
Other languages
English (en)
French (fr)
Inventor
刘向凤
林志远
林伟
芮华
Original Assignee
中兴通讯股份有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Priority claimed from CN202111511099.8A external-priority patent/CN116090326A/zh
Application filed by 中兴通讯股份有限公司 filed Critical 中兴通讯股份有限公司
Publication of WO2023071760A1 publication Critical patent/WO2023071760A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B7/00—Radio transmission systems, i.e. using radiation field
    • H04B7/02—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas
    • H04B7/04—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas
    • H04B7/06—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas at the transmitting station

Definitions

  • the present disclosure relates to the communication field, and in particular, to a method and device for dividing beam domains, a storage medium, and an electronic device.
  • MU-MIMO Multiple User Multiple Input Multiple Output
  • NR 5G New Radio
  • MU-MIMO utilizes spatial multiplexing technology to distinguish multi-stream data of multiple users
  • different users have different fading channel characteristics, and it is impossible to guarantee that the user channels are orthogonal to each other and do not interfere with each other. Therefore, it is necessary to select users with less correlation in the spatial domain for paired transmission as much as possible, so as to obtain the best spatial multiplexing gain. If the spatial domain can be divided into grids, some user pairing and user scheduling can be performed based on the grid mechanism, which will greatly reduce the search complexity and increase the multiplexing gain.
  • the grid-based mechanism can also reduce the implementation complexity and improve the detection performance.
  • the traditional method uses preset beams to perform rasterization operations on the spatial domain, resulting in low division accuracy, and no effective solution has been proposed yet.
  • Embodiments of the present disclosure provide a beam domain division method and device, a storage medium, and an electronic device, so as to at least solve the problem of low division accuracy caused by conventional methods using preset beams to perform rasterization operations on the spatial domain.
  • a beam domain division method including: acquiring a first set of measurement data; assigning a first label to each measurement data in the first set of measurement data, and through the The first label divides the beam domain corresponding to each measurement data in the spatial domain for each measurement data, wherein different measurement data in the first set of measurement data correspond to different first labels, and the beam The domain is used to indicate the grid that the spatial domain is divided in the process of meshing; the first group of measurement data with the first label is input into the neural network model for model training, and the beam domain division model is obtained; by The beam domain division model determines a second label corresponding to each measurement data in the second group of measurement data, so as to determine the location of each measurement data in the second group of measurement data in the spatial domain according to the second label.
  • the target beam domain in .
  • an apparatus for dividing beam domains including: an acquisition module configured to acquire a first set of measurement data; a first division module configured to obtain the first set of measurement data Each measurement data in is assigned a first label, and the beam domain corresponding to each measurement data is divided in the spatial domain by the first label, wherein the first set of measurement data Different measurement data in correspond to different first labels, and the beam domain is used to indicate the grid that the space domain divides in the process of meshing; the training module is set to use the first label with the first label A set of measurement data is input into the neural network model for model training to obtain a beam domain division model; the second division module is configured to determine the second label corresponding to each measurement data in the second group of measurement data through the beam domain division model , to determine a target beam domain in the spatial domain for each measurement data in the second group of measurement data according to the second label.
  • a computer-readable storage medium where a computer program is stored in the computer-readable storage medium, wherein the computer program is configured to execute the above-mentioned beam domain division method.
  • an electronic device including a memory, a processor, and a computer program stored in the memory and operable on the processor, wherein the processor executes the beam beam through the computer program. Domain division method.
  • the first set of measurement data is obtained, and each measurement data in the first set of measurement data is assigned a first label, and then the first label is used to divide each measurement data corresponding to each measurement data in the spatial domain.
  • the first group of measurement data with the first label is then input into the neural network model for model training to obtain the beam domain division model; the beam domain division model is used to determine the corresponding first group of measurement data in the second group of measurement data two labels, to determine the target beam domain in the spatial domain of each measurement data in the second group of measurement data according to the second label.
  • FIG. 1 is a block diagram of a hardware structure of a computer terminal according to a beam domain division method according to an embodiment of the present disclosure
  • FIG. 2 is a flowchart of a method for dividing beam domains according to an embodiment of the present disclosure
  • FIG. 3 is a schematic diagram of a beam domain division method based on a neural network according to an embodiment of the present disclosure
  • FIG. 4 is a schematic diagram of the corresponding relationship between beam domains and labels according to an embodiment of the present disclosure
  • FIG. 5 is a flowchart of a clustering-based beam domain division method according to an embodiment of the present disclosure
  • FIG. 6 is a schematic diagram (1) of a multi-user scenario constructed according to an embodiment of the present disclosure
  • Fig. 7 is a schematic diagram of a fully linked neural network used in an embodiment of the present disclosure.
  • FIG. 8 is a schematic diagram (2) of a multi-user scenario constructed according to an embodiment of the present disclosure.
  • FIG. 9 is a schematic diagram of a clustering-based beam domain division result according to an embodiment of the present disclosure.
  • Fig. 10 is a structural block diagram of an apparatus for dividing beam domains according to an embodiment of the present disclosure.
  • FIG. 1 is a block diagram of a hardware structure of a computer terminal according to a method for dividing beam domains according to an embodiment of the present disclosure.
  • the computer terminal can include one or more (only one is shown in Figure 1) processor 102 (processor 102 can include but not limited to microprocessor (Microprocessor Unit, MPU for short) or programmable logic A device (Programmable Logic device, PLD for short)) and a memory 104 for storing data.
  • processor 102 can include but not limited to microprocessor (Microprocessor Unit, MPU for short) or programmable logic A device (Programmable Logic device, PLD for short)
  • memory 104 for storing data.
  • the above-mentioned computer terminal may also include a transmission device 106 and an input and output device 108 for communication functions.
  • a transmission device 106 may also include a transmission device 106 and an input and output device 108 for communication functions.
  • the structure shown in FIG. 1 is only for illustration, and it does not limit the structure of the above computer terminal.
  • the computer terminal may also include more or less components than those shown in FIG. 1 , or have a different configuration with functions equivalent to those shown in FIG. 1 or more functions than those shown in FIG. 1 .
  • the memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as the computer program corresponding to the beam domain division method in the embodiment of the present disclosure, and the processor 102 runs the computer program stored in the memory 104, thereby Executing various functional applications and data processing is to realize the above-mentioned method.
  • the memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory.
  • the memory 104 may further include a memory that is remotely located relative to the processor 102, and these remote memories may be connected to a computer terminal through a network. Examples of the aforementioned networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.
  • the transmission device 106 is used to receive or transmit data via a network.
  • the specific example of the above-mentioned network may include a wireless network provided by the communication provider of the computer terminal.
  • the transmission device 106 includes a network interface controller (NIC for short), which can be connected to other network devices through a base station so as to communicate with the Internet.
  • the transmission device 106 may be a radio frequency (Radio Frequency, referred to as RF) module, which is used to communicate with the Internet in a wireless manner.
  • RF Radio Frequency
  • FIG. 2 is a flowchart of a method for dividing a beam domain according to an embodiment of the present disclosure. The process includes the following steps:
  • Step S202 acquiring the first set of measurement data
  • Step S204 assigning a first label to each measurement data in the first group of measurement data, and dividing the beam corresponding to each measurement data in the spatial domain for each measurement data through the first label domains, wherein different measurement data in the first group of measurement data correspond to different first labels, and the beam domain is used to indicate the grid that the spatial domain is divided into during the gridding process;
  • Step S206 inputting the first set of measurement data with the first label into the neural network model for model training to obtain a beam domain division model
  • Step S208 determine the second label corresponding to each measurement data in the second group of measurement data through the beam domain division model, so as to determine the position of each measurement data in the second group of measurement data according to the second label A target beam domain in the spatial domain.
  • the first set of measurement data is obtained, and each measurement data in the first set of measurement data is assigned a first label, and then the first label is used to divide each measurement data corresponding to each measurement data in the spatial domain.
  • the first group of measurement data with the first label is then input into the neural network model for model training to obtain the beam domain division model; the beam domain division model is used to determine the corresponding first group of measurement data in the second group of measurement data two labels, to determine the target beam domain in the spatial domain of each measurement data in the second group of measurement data according to the second label.
  • the above-mentioned determination of the second label corresponding to each measurement data in the second group of measurement data through the beam domain division model is realized in the following manner: in the beam domain division model, the first measurement data and the second When the similarity of the measurement data is greater than a preset threshold, the first label of the first measurement data is determined as the second label of the second measurement data, wherein the first group of measurement data includes: the first Measurement data, the second group of measurement data includes: second measurement data; it is determined in the beam domain division model that the similarity between the second measurement data and each measurement data in the first group of measurement data is less than the set In the case of the preset threshold, assign a second label to the second measurement data, where the second label is different from the first label of the first group of measurement data.
  • the present disclosure first selects a plurality of measurement data with low feature similarity from 100 data, assuming that it is 5 measurement data (equivalent to the first group of measurement data in the above-mentioned embodiment), for these 5
  • Each measurement data is assigned 5 different tags 1, 2, 3, 4, 5 (equivalent to the first tag in the above embodiment), and these 5 measurement data are divided into 5 different tags in the spatial domain according to the corresponding tags.
  • measurement data A is in beam domain 1
  • measurement data B is in beam domain 2
  • measurement data C is in beam domain 3.
  • Measure Data D is in beam domain 4
  • measurement data E is in beam domain 5 .
  • the corresponding relationship between the five measurement data and the beam domain is input into the neural network model for training, and the neural network model will learn the corresponding relationship between the characteristics of the measurement data and the beam domain, and finally obtain the beam domain division model.
  • each of the 95 measurement data is input into the beam domain division model , assuming that the measurement data F among the 95 measurement data is first input into the beam domain division model, and then the beam domain division model will compare the characteristics of the measurement data F with the characteristics of the measurement data A-E, when the characteristics of the measurement data F and the measurement When the feature similarity of each measurement data in the data A-E does not reach the preset threshold, the beam domain division model assigns a new label (equivalent to the second label in the above-mentioned embodiment) to the measurement data F, for example 6, And divide a new beam domain 6 in the spatial domain according to the new label, and then store the measurement data F in the beam domain 6 .
  • a new label equivalent to the second label in the above-mentioned embodiment
  • the beam domain division model When the similarity between the feature of the measurement data F and one of the measurement data A-E reaches the preset threshold, for example, the similarity between the feature of the measurement data F and the feature of the measurement data A reaches the preset threshold, the beam domain division model will output the measurement The label of data A, and then the measurement data F can be stored in beam domain 1. Furthermore, the beam domain division model can continuously divide the beam domain in the spatial domain according to the new measurement data.
  • determining the target beam domain of each measurement data in the second group of measurement data in the space domain according to the second label may be implemented in the following manner: The second tag in the beam field is compared with a plurality of the first tags one by one; if there is a first tag that is the same as the second tag in the current beam field to be divided among the plurality of first tags , acquiring the first measurement data corresponding to the first label that is the same as the second label of the current beam domain to be divided, and assigning the measurement data corresponding to the second label of the current beam domain to be divided to the first In the beam domain of the measurement data; in the case that there is no first label identical to the second label of the current beam domain to be divided among the multiple first labels, according to the second label of the current beam domain to be divided
  • the label is the measurement data corresponding to the second label of the current beam domain to be divided, and the target beam domain is divided in the spatial domain.
  • a plurality of second labels will be obtained, and then each of the plurality of second labels needs to be divided into the second label of the current beam domain to be divided and the first Multiple first tags in a set of measurement data are compared.
  • an example is used for illustration. There are 5 measurement data in the first measurement data, and there are 5 first labels, 1, 2, 3, 4, and 5, respectively. There are 2 measurement data in the second set of measurement data. After inputting into the beam domain division model, 2 second labels are obtained, assuming that the obtained labels are 5 and 6. Compare the 2 second tags with the 5 first tags.
  • one of the two second labels is the same as one of the five first labels, and then the test data corresponding to label 5 in the second set of measurement data is stored in the original label 5.
  • tag 6 none of the five first tags exists, and a new beam domain needs to be divided in the spatial domain according to tag 6, and the measurement data corresponding to tag 6 is stored in the beam domain 6 corresponding to the newly divided tag 6. middle.
  • the beam domain division model in order to let the beam domain division model know that the measurement data 6 is stored in the beam domain 6, it is necessary to input the relationship between the measurement data 6 and the beam domain 6 into the beam domain division model, so that the beam domain division model can be re- train. Specifically, it can be implemented in the following manner: adding the measurement data corresponding to the second label of the current beam domain to be divided into the beam domain division model for model training to obtain the target beam domain division model; dividing the beam domain into The model is replaced by the target beam domain partition model.
  • the beam domain division model it is necessary to perform feature extraction on each measurement data in the first group of measurement data with the first label, so as to obtain each A multi-dimensional feature data of measurement data; converting the format of the multi-dimensional feature data of each measurement data into a target format, wherein the target format is a format matched by the neural network model; The multi-dimensional feature data of each measurement data is input into the neural network model for model training to obtain a beam domain division model.
  • the beam domain division model allows the beam domain division model to know which measurement data corresponds to which beam domain.
  • the first set of measurement data with the first label is input into the neural network model for model training to obtain the beam domain division model, which is also realized in the following way: the first set of measurement data with the first label Divided into training set data and test set data; the training set data is input in the neural network model to carry out model training, and the first neural network model is obtained; the test set data is input in the first neural network model, Obtaining a first test result; when the first test result satisfies a preset condition, determining the first neural network model as a beam domain division model; when the first test result does not meet a preset condition Next, perform model training again by adjusting the parameter configuration of the neural network model.
  • the parameter configuration includes: the number of iterations, the number of neural network layers, the number of neurons, and the activation function.
  • data preprocessing may also be directly performed on the third set of measurement data to obtain a third set of measurement data after data preprocessing, wherein the third set of measurement data includes : the first group of measurement data, the second group of measurement data; the third group of measurement data after the data preprocessing is clustered through a clustering algorithm to obtain a third label; through the third label For each measurement data in the third group of measurement data, divide the beam domain corresponding to each measurement data in the spatial domain.
  • the first group of measurement data and the second group of measurement data can be directly used to determine the third label corresponding to each measurement data in the first group of measurement data and the second group of measurement data through a clustering algorithm, and through the third
  • the tag divides the corresponding beam domain in the space domain, and saves each measurement data into the divided beam domain.
  • the present disclosure is applicable to user selection and user scheduling when MUMIMO is used for data transmission in wireless communication systems, especially in Long Term Evolution (LTE for short) and 5G NR wireless communication systems, or in single-user multiple-input multiple-output ( Single User Multiple Input Multiple Output, referred to as SU-MIMO) transmission, the optimal beam selection field.
  • LTE Long Term Evolution
  • 5G NR wireless communication systems
  • SU-MIMO single-user multiple-input multiple-output
  • the embodiment of the present disclosure proposes a beam domain grid division method based on machine learning, so that more complex and real spatial domains can be constructed, making the grid division more accurate and reliable, mainly divided into beam domain division methods based on neural networks and Clustering-based Beam Domain Partitioning Method.
  • FIG. 3 is a schematic diagram of a beam domain division method based on a neural network according to an embodiment of the present disclosure, as shown in FIG. 3 .
  • the initial data collection and preprocessing modules include: data collection, screening processing, automatic label processing, and data preprocessing modules.
  • a period of user data (equivalent to the measurement data in the above-mentioned embodiment) is collected, including but not limited to the channel sounding reference signal (Sounding Reference Signal, referred to as SRS) channel response, signal and interference addition measured by the base station side.
  • SRS Sounding Reference Signal
  • SINR Signal to Interference plus Noise Ratio
  • PMI Precoding Matrix Indicator
  • CQI Channel Quality Indicator
  • DOA Direction of Arrival
  • the purpose of this module is to remove some abnormal data, so that the training sample data is as real and effective as possible user business data.
  • the screening criteria include but are not limited to filtering out data with a lower SINR according to the SINR.
  • the beam field grid table is preset.
  • the base station is covered by three sectors, the horizontal radiation range is plus or minus 60 degrees, and the downtilt is generally in the range of plus or minus 15 degrees, divided according to the horizontal and vertical steps of 5 degrees.
  • FIG. 4 is a schematic diagram of a corresponding relationship between beam domains and labels according to an embodiment of the present disclosure, and the two-dimensional preset beam domains are divided as shown in FIG. 4 below.
  • this module determines which grid belongs to the preset beam domain grid, and set the label, so that the training user data and the label establish a one-to-one correspondence.
  • the label in the preset beam field grid can be used, the label can also be reset, and the combination of vertical and horizontal angles can also be used as the label.
  • the purpose of this module is, firstly, when the data samples are not enough to cover the entire grid, choose to collect the scattered grid data of the beam domain grid as training samples as much as possible, as shown by the gray grid in Figure 4; Predicted new class calibration labels.
  • the multidimensional data features include:
  • SRS channel covariance matrix represents the user's vertical information and horizontal information
  • SINR It represents the coverage information and the area where the user is located
  • CQI represents channel quality information
  • DOA represents the user's vertical information and horizontal information
  • realizable multi-dimensional data features may be composed of one or more combinations of the above. Including but not limited to the following combinations:
  • the model training module based on the neural network is mainly divided into two stages: the initial training stage and the iterative training stage. in,
  • Initial training phase Divide the initially preprocessed multidimensional feature set data into two parts: training set and test set. Input the training set into the neural network for training, get the appropriate model parameters, and then use the test set to test the model and get the test results; continuously adjust and optimize the model parameters until a satisfactory test result is achieved, and save the trained Model parameters.
  • Iterative training phase If the matching probabilities of each beam domain detected online by the channel data of an unknown user are lower than a certain threshold, such as ⁇ , then it will be judged as a new beam domain, and the current user’s data will be saved and modeled. retraining and optimization. By analogy, the continuous expansion of the data sample set can continuously optimize the network model and detect more beam domains.
  • a certain threshold such as ⁇
  • the detection result of user data in an unknown location is identified as a new beam domain, the data is collected and labeled, and added to the training set for iterative training and model optimization.
  • the neural network model used in this application may be a convolutional neural network or a fully-connected neural network, but is not limited thereto, and may be any type of neural network.
  • the clustering-based beam domain division method in the embodiment of the present application mainly includes: data collection, data preprocessing, clustering, and beam domain division results.
  • a flow chart of the beam domain division method is shown in FIG. 5 . It should be noted that various clustering algorithms may be used in this embodiment of the present application.
  • machine learning is used for beam domain division, which may be supervised learning or unsupervised learning, and is not limited to neural network-based learning and clustering-based learning in the embodiments.
  • FIG. 6 is a schematic diagram (1) of a constructed multi-user scenario according to an embodiment of the present disclosure, (users are randomly distributed between y-100 and y-100).
  • the beam domain division method in this embodiment adopts four stages shown in FIG. 3 : initial data collection and preprocessing, model training module, online detection module, and feedback optimization module.
  • the initial data collection and preprocessing modules include: data collection, screening processing, automatic label processing, and data preprocessing modules.
  • user data is collected for a period of time, and what is collected in this embodiment is the SRS channel response and SINR value measured by the base station.
  • this module is to remove some abnormal data, so that the training sample data is as real and effective as possible user business data.
  • the data whose SINR is lower than -10dB is filtered out.
  • the beam field grid table is preset.
  • the base station is covered by three sectors, the horizontal radiation range is plus or minus 60 degrees, and the downtilt is generally in the range of plus or minus 15 degrees, divided according to the horizontal and vertical steps of 5 degrees.
  • the two-dimensional preset beam domain is divided as shown in FIG. 4 .
  • the vertical covariance matrix R v and the horizontal covariance matrix R h are calculated according to the SRS channel estimation value and the antenna array mode. After transforming R v to the angle domain, search for the peak in the angle domain, and calculate the vertical angle ⁇ according to the peak position. After transforming Rh into the angle domain, search for the peak in the angle domain, and calculate the horizontal angle according to the peak position
  • the labels in the preset beam domain grid are used, that is, the numbers with background colors shown in FIG. 4 .
  • the first column of the SRS channel covariance matrix is used.
  • the SRS channel covariance matrix Rx E[H*H H ], where H represents the SRS channel response, and () H represents the conjugate transpose.
  • the input data format of the neural network or clustering algorithm is the first column vector of Rx. Since the currently applied machine learning requires the input to be a real number and the input value needs to be normalized, the format of the constructed input feature data is:
  • ); normRx Rx1/k mod; the number of rows of input feature data is the number of samples, and the number of columns is the number of features.
  • the learning and training module based on the neural network is mainly divided into two stages: the initial training stage and the iterative training stage.
  • Initial training phase Divide the initially preprocessed multidimensional feature set data into two parts: training set and test set. Input the training set into the neural network for training, get the appropriate model parameters, and then use the test set to test the model and get the test results; continuously adjust and optimize the model parameters until a satisfactory test result is achieved, and save the trained Model parameters.
  • the same data preprocessing is performed on the channel data of users with unknown locations to obtain a multi-dimensional feature set, which is input into the trained neural network for online detection, and the preset matching probability is given, and the final detection result is the one with the highest probability.
  • the prediction accuracy rate of spatial beam domain division based on the neural network is over 95%.
  • the data will be collected and labeled, and added to the training set for iterative training and model optimization.
  • the neural network model used in this embodiment is a fully linked four-layer neural network, an input layer, two hidden layers, and an output layer.
  • the optimizer used is stochastic gradient descent (Stochastic Gradient Descent, referred to as SGD), and the activation function is Logistic .
  • Fig. 7 is a schematic diagram of a fully connected neural network used in an embodiment of the present disclosure.
  • Fig. 8 is a schematic diagram (2) of a multi-user scenario constructed in an embodiment of the present disclosure. There may also be multiple input data feature sets based on the clustering algorithm, and the first column of the channel covariance matrix is used in this embodiment.
  • the SRS channel covariance matrix Rx E[H*H H ], where H represents the SRS channel response, and () H represents the conjugate transpose. Since the currently applied machine learning requires the input to be a real number and the input value needs to be normalized, the format of the constructed input feature data is:
  • ); normRx Rx1/k mod; the number of rows of input feature data is the number of samples, and the number of columns is the number of features.
  • sklearn.cluster.KMeans and Bio.Cluste.kcluster modules are used to realize the clustering algorithm, so as to realize the division of beam domains of different users.
  • Fig. 9 is a schematic diagram of beam domain division results based on clustering in an embodiment of the present disclosure (the horizontal axis is the number of samples, the vertical axis is the label number, the gray level is the cluster identification, the same gray level represents the same class, and the division accuracy is 100%).
  • the clustering method and the neural network-based method can be used in combination, and the neural network can be further used for more accurate beam domain division based on the prior results of the clustering.
  • the beneficial effects of the present disclosure are as follows: (1) In the MU-MIMO technology in the wireless communication system, space beam domain division is performed based on machine learning, which is more suitable for the actual changeable and complex environment, and has Help improve the matching success rate. (2) On-line detection of the user's belonging beam domain, iterative training and detection according to the detection results, and continuous expansion of the training data set are easy to implement and can improve detection performance. (3) Based on the division of spatial beam domains and the statistics of spatial characteristics, in the application of MU-MIMO technology, it helps to reduce the complexity of user pairing, expand the range of user pairing options, and easily select the optimal combination to obtain the best pairing gain . (4) By dividing the space beam domain, it helps to reduce the complexity of user scheduling in the application of MU-MIMO technology. (5) The multi-dimensional feature set extracted in the present disclosure is easy to calculate and easy to realize by hardware.
  • the above-mentioned technical solution in the embodiments of the present application can perform spatial beam domain division and perform home beam domain detection on users.
  • MU-MIMO technology it can be applied to multi-user pairing and multi-user scheduling to improve MU-MIMO transmission performance.
  • SU transmission it helps in the selection of the optimal beam.
  • the positioning system the positioning area can be reduced to reduce the complexity of positioning.
  • spatial rasterization helps to reduce complexity and improve detection performance.
  • the embodiment of the present application can be based on machine learning user selection, user scheduling, and multi-user clustering judgment; in the SU-MIMO scenario, it will involve the spatially optimal beam domain selection; in other scenarios , For example, in the positioning system, in the detection system, and in the monitoring system, the division of space domains will also be involved.
  • the method according to the above embodiments can be implemented by means of software plus a necessary general-purpose hardware platform, and of course also by hardware, but in many cases the former is Better implementation.
  • the technical solution of the present disclosure can be embodied in the form of a software product in essence or the part that contributes to the prior art, and the computer software product is stored in a storage medium (such as ROM/RAM, disk, CD) contains several instructions to make a terminal device (which may be a mobile phone, a computer, a server, or a network device, etc.) execute the methods of various embodiments of the present disclosure.
  • a device for dividing beam domains is also provided, and the device is used to implement the above embodiments and preferred implementation manners, and what has been explained will not be repeated here.
  • the term "module” may be a combination of software and/or hardware that realizes a predetermined function.
  • the devices described in the following embodiments are preferably implemented in software, implementations in hardware, or a combination of software and hardware are also possible and contemplated.
  • Fig. 10 is a structural block diagram of an apparatus for dividing a beam domain according to an embodiment of the present disclosure, the apparatus includes:
  • the acquisition module 10 is configured to acquire the first set of measurement data
  • the first division module 12 is configured to assign a first label to each measurement data in the first group of measurement data, and divide each measurement data in the spatial domain by using the first label.
  • the beam field corresponding to the measurement data wherein different measurement data in the first group of measurement data correspond to different first labels, and the beam field is used to indicate the grid that the space domain is divided into during the gridding process grid;
  • the training module 14 is configured to input the first set of measurement data with the first label into the neural network model for model training to obtain a beam domain division model;
  • the second division module 16 is configured to determine the second label corresponding to each measurement data in the second group of measurement data through the beam domain division model, so as to determine the second label in the second group of measurement data according to the second label.
  • the first set of measurement data is obtained, and each measurement data in the first set of measurement data is assigned a first label, and then the first label is used to divide each measurement data corresponding to each measurement data in the spatial domain.
  • the first group of measurement data with the first label is then input into the neural network model for model training to obtain the beam domain division model; the beam domain division model is used to determine the corresponding first group of measurement data in the second group of measurement data two labels, to determine the target beam domain in the spatial domain of each measurement data in the second group of measurement data according to the second label.
  • the second division module 16 is also configured to divide the second measurement data of the first measurement data into A tag is determined as the second tag of the second measurement data, wherein the first set of measurement data includes: first measurement data, and the second set of measurement data includes: second measurement data;
  • assigning a second label to the second measurement data assigning a second label to the second measurement data, the A second label is different from said first label of said first set of measurements.
  • the second division module 16 is also configured to compare the second label of the current beam domain to be divided among the second labels with a plurality of the first labels one by one; In the case where there is a first label identical to the second label of the current beam domain to be divided, acquire the first measurement data corresponding to the first label identical to the second label of the current beam domain to be divided, and Allocating the measurement data corresponding to the second label of the current beam domain to be divided into the beam domain of the first measurement data; there is no second label corresponding to the current beam domain to be divided among the plurality of first labels In a case where the two labels are the same as the first label, divide the target beam domain in the spatial domain according to the second label of the current beam domain to be divided for the measurement data corresponding to the second label of the current beam domain to be divided. Wherein, the number of the first tags is multiple, and the number of the first tags is the same as the number of measurement data in the first set of measurement data.
  • the training module 14 is also configured to add the measurement data corresponding to the second label of the current beam domain to be divided to the beam domain division model for model training to obtain a target beam domain division model;
  • the division model is replaced by the target beam domain division model.
  • the training module 14 is also configured to perform feature extraction on each measurement data in the first group of measurement data with the first label, so as to obtain the multi-dimensional feature data of each measurement data;
  • the format of the multidimensional feature data of a measurement data is converted into a target format, wherein the target format is a format matched by the neural network model;
  • the multidimensional feature data of each measurement data with the target format is input into the Model training is carried out in the above neural network model, and the beam domain division model is obtained.
  • the training module 14 is also configured to divide the first group of measurement data with the first label into training set data and test set data; input the training set data into the neural network model for model training, obtaining a first neural network model; inputting the test set data into the first neural network model to obtain a first test result; when the first test result satisfies a preset condition, the first neural network
  • the network model is determined to be a beam domain division model; if the first test result does not meet the preset condition, model training is performed again by adjusting the parameter configuration of the neural network model.
  • the training module 14 is also configured to adjust the parameter configuration of the neural network model, and add the training set data to the adjusted neural network model for training to obtain a second neural network model; Inputting the collected data into the second neural network model to obtain a second test result; if the second test result satisfies a preset condition, determining the second neural network model as a beam domain division model.
  • the first division module 12 is also configured to perform data preprocessing on the third group of measurement data to obtain a third group of measurement data after data preprocessing, wherein the third group of measurement data includes: the first A group of measurement data, the second group of measurement data; the third group of measurement data after the data preprocessing is clustered through a clustering algorithm to obtain a third label; through the third label for the third
  • Each measurement data in the set of measurement data divides the beam domain corresponding to each measurement data in the spatial domain.
  • Embodiments of the present disclosure also provide a computer-readable storage medium, in which a computer program is stored, wherein the computer program is configured to execute the steps in any one of the above method embodiments when running.
  • the above-mentioned storage medium may be configured to store a computer program for performing the following steps:
  • the above-mentioned computer-readable storage medium may include but not limited to: U disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM) , mobile hard disk, magnetic disk or optical disk and other media that can store computer programs.
  • ROM read-only memory
  • RAM random access memory
  • mobile hard disk magnetic disk or optical disk and other media that can store computer programs.
  • Embodiments of the present disclosure also provide an electronic device, including a memory and a processor, where a computer program is stored in the memory, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.
  • the above-mentioned processor may be configured to execute the following steps through a computer program:
  • the electronic device may further include a transmission device and an input and output device, wherein the transmission device is connected to the processor, and the input and output device is connected to the processor.
  • each module or each step of the above-mentioned disclosure can be realized by a general-purpose computing device, and they can be concentrated on a single computing device, or distributed in a network composed of multiple computing devices In fact, they can be implemented in program code executable by a computing device, and thus, they can be stored in a storage device to be executed by a computing device, and in some cases, can be executed in an order different from that shown here. Or described steps, or they are fabricated into individual integrated circuit modules, or multiple modules or steps among them are fabricated into a single integrated circuit module for implementation. As such, the present disclosure is not limited to any specific combination of hardware and software.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Molecular Biology (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Radio Transmission System (AREA)

Abstract

本公开提供了一种波束域的划分方法及装置、存储介质及电子装置，其中，上述方法包括：获取第一组测量数据；为所述第一组测量数据中的每一测量数据分配第一标签，并通过所述第一标签为所述每一测量数据在空间域中划分所述每一测量数据对应的波束域，其中，所述第一组测量数据中的不同测量数据对应不同的第一标签，所述波束域用于指示所述空间域在进行网格化的过程中划分的栅格；将具有所述第一标签的第一组测量数据输入神经网络模型中进行模型训练，得到波束域划分模型；通过所述波束域划分模型确定第二组测量数据中的每一测量数据对应的第二标签，以根据所述第二标签确定所述第二组测量数据中的每一测量数据在所述空间域中的目标波束域。

Description

波束域的划分方法及装置、存储介质及电子装置
本公开要求于2021年10月29日提交中国专利局、申请号为202111277675.7、发明名称“波束域的划分方法及装置、存储介质及电子装置技”的中国专利申请的优先权，以及要求于2021年12月10日提交中国专利局、申请号为202111511099.8、发明名称“波束域的划分方法及装置、存储介质及电子装置”的中国专利申请的优先权，其全部内容通过引用结合在本公开中。
技术领域
本公开涉及通信领域，具体而言，涉及一种波束域的划分方法及装置、存储介质及电子装置。
背景技术
在无线通信系统中，多用户多入多出(Multiple User Multiple Input MultipleOutput，简称为MU-MIMO)在相同的时频资源上利用空间复用技术进行多个用户的数据传输，显著提高了系统的吞吐量。随着人们对传输速率和服务质量越来越高的要求，在当下的5G新无线(New Radio，简称为NR)和将来的6G中，MU-MIMO将会受到越来越多的关注。
然而，MU-MIMO利用空间复用技术区分多个用户的多流数据时，理论是需要各个用户的信道具有较好的正交性。但是，实际的传输环境中，根据无线通信传输的空间多样性、随机性，不同的用户则有着不同的衰落信道特性，是无法保证用户信道相互正交，相互无干扰的。因此需要尽可能的选择空间域上相关度较小的用户进行配对传输，以获得最好的空间复用增益。如果能将空间域进行一些网格化的划分，基于栅格化机制进行一些用户配对，用户调度等等，将极大的降低搜索复杂度，提高复用增益。
另外，在定位系统中或探测系统中，基于栅格化机制同样可以降低实现复杂度，提高检测性能。
在当前的一些文献中，有对空间的栅格化操作，大都是基于预置波束进行划分，或者基于测量参数进行划分，划分精度不够高，或者划分不够准确，或者重叠划分，使用时需要搜索，复杂度高。
针对相关技术中，传统方法采用预置波束对空间域进行栅格化操作，导致划分精度较低的问题，目前尚未提出有效的解决方案。
因此，有必要对相关技术予以改良以克服相关技术中的所述缺陷。
发明内容
本公开实施例提供了一种波束域的划分方法及装置、存储介质及电子装置，以至少解决传统方法采用预置波束对空间域进行栅格化操作，导致划分精度较低的问题。
根据本公开实施例的一方面，提供一种波束域的划分方法，包括：获取第一组测量数据；为所述第一组测量数据中的每一测量数据分配第一标签，并通过所述第一标签为所述每一测量数据在空间域中划分所述每一测量数据对应的波束域，其中，所述第一组测量数据中的不同测量数据对应不同的第一标签，所述波束域用于指示所述空间域在进行网格化的过程中划 分的栅格；将具有所述第一标签的第一组测量数据输入神经网络模型中进行模型训练，得到波束域划分模型；通过所述波束域划分模型确定第二组测量数据中的每一测量数据对应的第二标签，以根据所述第二标签确定所述第二组测量数据中的每一测量数据在所述空间域中的目标波束域。
根据本公开实施例的另一方面，还提供了一种波束域的划分装置，包括：获取模块，设置为获取第一组测量数据；第一划分模块，设置为为所述第一组测量数据中的每一测量数据分配第一标签，并通过所述第一标签为所述每一测量数据在空间域中划分所述每一测量数据对应的波束域，其中，所述第一组测量数据中的不同测量数据对应不同的第一标签，所述波束域用于指示所述空间域在进行网格化的过程中划分的栅格；训练模块，设置为将具有所述第一标签的第一组测量数据输入神经网络模型中进行模型训练，得到波束域划分模型；第二划分模块，设置为通过所述波束域划分模型确定第二组测量数据中的每一测量数据对应的第二标签，以根据所述第二标签确定所述第二组测量数据中的每一测量数据在所述空间域中的目标波束域。
根据本公开实施例的又一方面，还提供了一种计算机可读的存储介质，该计算机可读的存储介质中存储有计算机程序，其中，该计算机程序被设置为运行时执行上述波束域的划分方法。
根据本公开实施例的又一方面，还提供了一种电子装置，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，其中，上述处理器通过计算机程序执行上述波束域的划分方法。
通过本公开，获取第一组测量数据，并为第一组测量数据中的每一测量数据分配第一标签，进而通过第一标签为每一测量数据在空间域中划分每一测量数据对应的波束域，随后将具有第一标签的第一组测量数据输入神经网络模型中进行模型训练，得到波束域划分模型；通过波束域划分模型确定第二组测量数据中的每一测量数据对应的第二标签，以根据第二标签确定第二组测量数据中的每一测量数据在空间域中的目标波束域。
附图说明
此处所说明的附图用来提供对本公开的进一步理解，构成本申请的一部分，本公开的示例性实施例及其说明用于解释本公开，并不构成对本公开的不当限定。在附图中：
图1是本公开实施例的波束域的划分方法的计算机终端的硬件结构框图；
图2是本公开实施例的波束域的划分方法的流程图；
图3是根据本公开实施例的基于神经网络的波束域划分方法的示意图；
图4是根据本公开实施例的波束域与标签的对应关系示意图；
图5是根据本公开实施例的基于聚类的波束域划分方法的流程图；
图6是根据本公开实施例的构建的多用户场景示意图(一)；
图7是根据本公开实施例中采用的全链接神经网络示意图；
图8是根据本公开实施例中构建的多用户场景示意图(二)；
图9是根据本公开实施例中基于聚类的波束域划分结果示意图；
图10是根据本公开实施例的波束域的划分装置的结构框图。
具体实施方式
为了使本技术领域的人员更好地理解本公开方案，下面将结合本公开实施例中的附图，对本公开实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本公开一部分的实施例，而不是全部的实施例。基于本公开中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都应当属于本公开保护的范围。
需要说明的是，本公开的说明书和权利要求书及上述附图中的术语“第一”、“第二”等是用于区别类似的对象，而不必用于描述特定的顺序或先后次序。应该理解这样使用的数据在适当情况下可以互换，以便这里描述的本公开的实施例能够以除了在这里图示或描述的那些以外的顺序实施。此外，术语“包括”和“具有”以及他们的任何变形，意图在于覆盖不排他的包含，例如，包含了一系列步骤或单元的过程、方法、系统、产品或设备不必限于清楚地列出的那些步骤或单元，而是可包括没有清楚地列出的或对于这些过程、方法、产品或设备固有的其它步骤或单元。
本申请实施例中所提供的方法实施例可以在计算机终端或者类似的运算装置中执行。以运行在计算机终端上为例，图1是本公开实施例的波束域的划分方法的计算机终端的硬件结构框图。如图1所示，计算机终端可以包括一个或多个(图1中仅示出一个)处理器102(处理器102可以包括但不限于微处理器(Microprocessor Unit，简称是MPU)或可编程逻辑器件(Programmable Logic device，简称是PLD))和用于存储数据的存储器104，在一个示例性实施例中，上述计算机终端还可以包括用于通信功能的传输设备106以及输入输出设备108。本领域普通技术人员可以理解，图1所示的结构仅为示意，其并不对上述计算机终端的结构造成限定。例如，计算机终端还可包括比图1中所示更多或者更少的组件，或者具有与图1所示等同功能或比图1所示功能更多的不同的配置。
存储器104可用于存储计算机程序，例如，应用软件的软件程序以及模块，如本公开实施例中的波束域的划分方法对应的计算机程序，处理器102通过运行存储在存储器104内的计算机程序，从而执行各种功能应用以及数据处理，即实现上述的方法。存储器104可包括高速随机存储器，还可包括非易失性存储器，如一个或者多个磁性存储装置、闪存、或者其他非易失性固态存储器。在一些实例中，存储器104可进一步包括相对于处理器102远程设置的存储器，这些远程存储器可以通过网络连接至计算机终端。上述网络的实例包括但不限于互联网、企业内部网、局域网、移动通信网及其组合。
传输装置106用于经由一个网络接收或者发送数据。上述的网络具体实例可包括计算机终端的通信供应商提供的无线网络。在一个实例中，传输装置106包括一个网络适配器(Network Interface Controller，简称为NIC)，其可通过基站与其他网络设备相连从而可与互联网进行通讯。在一个实例中，传输装置106可以为射频(Radio Frequency，简称为RF)模块，其用于通过无线方式与互联网进行通讯。
在本实施例中提供了一种波束域的划分方法，图2是根据本公开实施例的波束域的划分方法的流程图，该流程包括如下步骤：
步骤S202，获取第一组测量数据；
步骤S204，为所述第一组测量数据中的每一测量数据分配第一标签，并通过所述第一标签为所述每一测量数据在空间域中划分所述每一测量数据对应的波束域，其中，所述第一组测量数据中的不同测量数据对应不同的第一标签，所述波束域用于指示所述空间域在进行网 格化的过程中划分的栅格；
步骤S206，将具有所述第一标签的第一组测量数据输入神经网络模型中进行模型训练，得到波束域划分模型；
步骤S208，通过所述波束域划分模型确定第二组测量数据中的每一测量数据对应的第二标签，以根据所述第二标签确定所述第二组测量数据中的每一测量数据在所述空间域中的目标波束域。
通过上述步骤，获取第一组测量数据，并为第一组测量数据中的每一测量数据分配第一标签，进而通过第一标签为每一测量数据在空间域中划分每一测量数据对应的波束域，随后将具有第一标签的第一组测量数据输入神经网络模型中进行模型训练，得到波束域划分模型；通过波束域划分模型确定第二组测量数据中的每一测量数据对应的第二标签，以根据第二标签确定第二组测量数据中的每一测量数据在空间域中的目标波束域。采用上述技术方案，解决了传统方法采用预置波束对空间域进行栅格化操作，导致划分精度较低的问题。进而采用上述技术方案，利用神经网络进行波束域栅格划分，能够学习更复杂的更真实的空间域，使得栅格划分更准确可靠。
进一步地，上述通过所述波束域划分模型确定第二组测量数据中的每一测量数据对应的第二标签，通过以下方式实现：在所述波束域划分模型中确定第一测量数据与第二测量数据的相似度大于预设阈值的情况下，将所述第一测量数据的第一标签确定为所述第二测量数据的第二标签，其中，所述第一组测量数据包括：第一测量数据，所述第二组测量数据包括：第二测量数据；在所述波束域划分模型中确定第二测量数据与所述第一组测量数据中的每一测量数据的相似度都小于所述预设阈值的情况下，为所述第二测量数据分配第二标签，其中，所述第二标签与所述第一组测量数据的第一标签不同。
为了更好的理解上述步骤，以下举例进行说明，假设现在有100个测量数据，需要放入空间域中的栅格中，传统的方法先将空间域划分成固定数量的多个栅格，需要说明的是，一个栅格为一个波束域，不同的栅格存放不同特征的测量数据，进而将100个测量数据根据特征的不同，存放在空间域中固定数量的栅格中。但此种方法，由于空间域已经预先划分好了，所以划分的精度不够高，可能会存在相似度不是很高的两组测量数据放在同一个栅格，即同一个波束域中。
为了解决上述问题，本公开先从100个数据中挑选出特征相似度不高的多个测量数据，假设为5个测量数据(相当于上述实施例中的第一组测量数据)，为这5个测量数据分配5个不同的标签1、2、3、4、5(相当于上述实施例中的第一标签)，将这5个测量数据分别按照对应的标签划分在空间域中5个不同的波束域，将这5个测量数据存放入这5个对应的波束域中，例如测量数据A在波束域1中，测量数据B在波束域2中，测量数据C在波束域3中，测量数据D在波束域4中，测量数据E在波束域5中。进而将这五个测量数据与波束域的对应关系输入到神经网络模型中进行训练，神经网络模型会学习到测量数据的特征和波束域的对应关系，最后得到波束域划分模型。
需要说明的是，100个数据中还剩下95个数据(相当于上述实施例中的第二组测量数据)，将这95个测量数据中的每一个测量数据都输入到波束域划分模型中，假设先将这95个测量数据中的测量数据F输入到波束域划分模型，进而波束域划分模型会将测量数据F的特征与测量数据A-E的特征进行对比，当测量数据F的特征与测量数据A-E中每一个测量数据的特 征相似度都没有达到预设阈值的情况下，波束域划分模型为测量数据F分配一个新的标签(相当于上述实施例中的第二标签)，例如6，并根据这个新的标签在空间域中划分新的波束域6，进而可以将测量数据F存放在波束域6中。当测量数据F的特征与测量数据A-E中其中一个测量数据的特征相似度达到预设阈值，例如测量数据F的特征与测量数据A的特征相似度达到预设阈值，波束域划分模型会输出测量数据A的标签，进而可以将测量数据F存放在波束域1中。进而波束域划分模型可以不断的根据新来的测量数据，在空间域中划分波束域。
进一步的，根据所述第二标签确定所述第二组测量数据中的每一测量数据在所述空间域中的目标波束域，可以通过以下方式实现：将所述第二标签中当前待划分波束域的第二标签与多个所述第一标签一一进行比对；在多个所述第一标签中存在与所述当前待划分波束域的第二标签相同的第一标签的情况下，获取与所述当前待划分波束域的第二标签相同的第一标签所对应的第一测量数据，并将所述当前待划分波束域的第二标签对应的测量数据分配至所述第一测量数据的波束域中；在多个所述第一标签中不存在与所述当前待划分波束域的第二标签相同的第一标签的情况下，根据所述当前待划分波束域的第二标签为所述当前待划分波束域的第二标签对应的测量数据在所述空间域中划分目标波束域。
也就是说，将第二组测量数据输入到波束域划分模型以后，会得到多个第二标签，进而需要将这多个第二标签中的每一个当前待划分波束域的第二标签与第一组测量数据中的多个第一标签进行比对。为了更好的理解，举例进行说明，第一测量数据中有5个测量数据，分别有5个第一标签，1、2、3、4、5。第二组测量数据中有2个测量数据，输入到波束域划分模型以后，得到2个第二标签，假设得到的标签为5、6。将这2个第二标签与5个第一标签进行对比。从对比结果可知，这2个第二标签中存在一个标签与5个第一标签中的一个标签相同，进而将第二组测量数据中标签5对应的测试数据，存放在原有的标签5对应的波束域5中。而对于标签6，在这5个第一标签中都没有，进而需要根据标签6在空间域中新划分一个波束域，将标签6对应的测量数据存放在新划分的标签6对应的波束域6中。
需要说明的是，为了让波束域划分模型知道测量数据6是存放在波束域6中的，进而需要将测量数据6对应波束域6的关系输入到波束域划分模型中，让波束域划分模型重新训练。具体的，可以通过以下方式实现：将所述当前待划分波束域的第二标签对应的测量数据添加到所述波束域划分模型进行模型训练，得到目标波束域划分模型；将所述波束域划分模型替换为所述目标波束域划分模型。
进一步地，为了更好的理解波束域划分模型是如何训练出来的，具体的，需要对具有所述第一标签的第一组测量数据中的每一测量数据进行特征提取，以得到所述每一测量数据的多维特征数据；将所述每一测量数据的多维特征数据的格式转化为目标格式，其中，所述目标格式为所述神经网络模型所匹配的格式；将具有所述目标格式的所述每一测量数据的多维特征数据输入所述神经网络模型中进行模型训练，得到波束域划分模型。
也就是说，需要从第一组测量数据中的每一个测量数据中提取对应的多维特征数据，并且将多维特征数据与第一标签的对应关系输入到神经网络模型中进行训练，进而才可以确定波束域划分模型，让波束域划分模型知道哪一个测量数据对应哪一个波束域。
进一步地，将具有所述第一标签的第一组测量数据输入神经网络模型中进行模型训练，得到波束域划分模型，还通过以下方式实现：将具有所述第一标签的第一组测量数据划分为训练集数据和测试集数据；将所述训练集数据输入所述神经网络模型中进行模型训练，得到 第一神经网络模型；将所述测试集数据输入所述第一神经网络模型中，得到第一测试结果；在所述第一测试结果满足预设条件的情况下，将所述第一神经网络模型确定为波束域划分模型；在所述第一测试结果不满足预设条件的情况下，通过调整所述神经网络模型的参数配置再次进行模型训练。
需要说明的是，通过调整所述神经网络模型的参数配置再次进行模型训练，需要调整神经网络模型的参数配置，并将训练集数据添加到调整后的神经网络模型中进行训练，得到第二神经网络模型；将所述测试集数据输入所述第二神经网络模型中，得到第二测试结果；并在第二测试结果满足预设条件的情况下，将所述第二神经网络模型确定为波束域划分模型。
需要说明的是，参数配置包括：迭代次数、神经网络层数、神经元数，激活函数。需要说明的是，在一个可选的实施例中，还可以直接对第三组测量数据进行数据预处理，得到数据预处理后的第三组测量数据，其中，所述第三组测量数据包括：所述第一组测量数据，所述第二组测量数据；将所述数据预处理后的第三组测量数据通过聚类算法进行聚类处理，得到第三标签；通过所述第三标签为所述第三组测量数据中的每一测量数据在空间域中划分所述每一测量数据对应的波束域。
也就是说，可以直接将第一组测量数据与第二组测量数据一起通过聚类算法确定第一组测量数据和第二组测量数据中每一个测量数据对应的第三标签，并通过第三标签在空间域中划分对应的波束域，将每一个测量数据保存进划分好的波束域中。
显然，上述所描述的实施例仅仅是本公开一部分的实施例，而不是全部的实施例。为了更好的理解上述波束域的划分方法，以下结合实施例对上述过程进行说明，但不用于限定本公开实施例的技术方案，具体地：
需要说明的是。本公开适用于无线通信系统，尤其是长期演进(Long Term Evolution，简称为LTE)和5G NR无线通信系统中利用MUMIMO进行数据传输时的用户选择、用户调度，或者在单用户多入多出(Single User Multiple Input Multiple Output，简称为SU-MIMO)传输中，最优波束的选择领域。
本公开实施例提出了一种基于机器学习的波束域栅格划分方法，使得可以构建更复杂更真实的空间域，使得栅格划分更准确可靠，主要分为基于神经网络的波束域划分方法和基于聚类的波束域划分方法。
其中，基于神经网络的波束域划分方法主要分为四个阶段：数据初始收集与预处理、模型训练模块、在线检测模块、反馈优化模块。图3是根据本公开实施例的基于神经网络的波束域划分方法的示意图，如图3所示。
一、数据初始收集与预处理
数据初始收集与预处理模块包括：数据收集、筛选处理、自动标签处理、数据预处理模块。
a.数据收集
在初始阶段，收集一段时间的用户数据(相当于上述实施例中的测量数据)，包含但不限于基站侧测量的信道探测参考信号(Sounding Reference Signal，简称为SRS)信道响应、信号与干扰加噪声比(Signal to Interference plus Noise Ratio，简称为SINR)值，预编码矩阵指示(Precoding Matrix Indicator，简称为PMI)值，信道质量指示(Channel Quality Indicator，简称为CQI)值，波达方向(Direction of Arrival，简称为DOA)值 等等，根据实际的机器学习的输入数据设计而定。
b.筛选处理
本模块的目的主要是去除掉一些异常数据，使得训练样本数据尽可能的是真实有效的用户业务数据。筛选准则包含但不限于根据SINR，将SINR较低的数据滤除掉。
c.自动标签处理
根据实际基站的覆盖范围，预置波束域栅格表。一般基站是三扇区覆盖，水平辐射范围为正负60度，下倾角一般在正负15度范围，按照水平和垂直5度的步长进行划分。图4是根据本公开实施例的波束域与标签的对应关系示意图，则二维的预置波束域划分为如下图4所示。
如果有上报的DOA测量量，则利用上报的DOA值，如果没有，根据SRS信道估计值及天线阵列模式，计算垂直向和水平向的协方差矩阵，将协方差矩阵变换到角度域估计垂直向和水平向角度。根据估计出的垂直和水平向角度，判断归属预置波束域栅格的哪个栅格，进行标签的设置，这样训练用户数据与标签就建立起一一对应关系。可以利用预置波束域栅格里的标签，也可以重新设置标签，也可以利用垂直和水平向角度的组合作为标签。本模块的目的，一是在数据样本不够覆盖整个栅格时，尽可能选择收集波束域栅格零散分布的栅格数据作为训练样本，如图4中灰色的栅格所示；二是在给预测的新类标定标签。
d.数据预处理
提取多维数据特征，将数据预处理为机器学习所需要的数据格式。其中多维数据特征包括：
(1)SRS信道协方差矩阵：表征了用户的垂直向信息和水平向信息；
(2)SINR：表征了覆盖信息，用户所处的区域位置；
(3)PMI：表征了用户的波束方向信息；
(4)CQI：表征了信道质量信息；
(5)DOA：表征了用户的垂直向信息和水平向信息；
需要说明的是，可实现的多维数据特征可由上述的一种或多种组合构成。包含但不限于以下组合：
(1)利用SRS信道协方差矩阵，包含但不限于以下几种格式：
1)SRS信道协方差矩阵的第一列。
2)SRS信道垂直向协方差矩阵的第一列、SRS信道水平向协方差矩阵的第一列。或者垂直向和水平向可以分开训练检测，之后将检测结果进行并集处理。
3)垂直向协方差矩阵的能量谱，水平向协方差矩阵的能量谱。
4)垂直向协方差矩阵的能量谱，垂直向能量谱的峰值位置，水平向协方差矩阵的能量谱，水平向能量谱的峰值位置。
5)赋形权值向量最大能量方向的垂直向分量的能量谱，赋形权值向量最大能量方向的水平向分量的能量谱。
6)赋形权值向量最大能量方向的垂直向分量的能量谱，垂直向能量谱的峰值位置，赋形权值向量最大能量方向的水平向分量的能量谱，水平向能量谱的峰值位置。
(2)SRS信道协方差矩阵与SINR组合，可学习预测不同覆盖下的空间波束域；
(3)SRS信道协方差矩阵与CQI组合，可学习预测不同覆盖下的空间波束域；
(4)SRS信道协方差矩阵与PMI组合，可提高学习预测准确度；
(5)SRS信道协方差矩阵与DOA组合，可提高学习预测准确度。
二、模型训练模块
基于神经网络的模型训练模块主要分两个阶段：初始训练阶段、迭代训练阶段。其中，
初始训练阶段：将初始预处理好的多维特征集数据分为两部分：训练集和测试集。把训练集输入到神经网络中进行训练，得到合适的模型参数，然后利用测试集进行模型的测试，得到测试结果；不断调整和优化模型参数，一直达到比较满意的测试结果为止，保存训练好的模型参数。
迭代训练阶段：如果未知位置用户的信道数据在线检测出来的归属各个波束域的匹配概率都低于某一个阈值，比如β，那么会被判为新的波束域，保存当前用户的数据，进行模型的重新训练和优化。以此类推，这样不断扩大数据样本集，能够使网络模型不断优化，检测出更多的波束域。
三、在线检测模块
将未知位置的用户的数据，进行同样的数据预处理，获取多维特征集，输入到训练好的神经网络(相当于上述波束域划分模型)中进行在线检测，给出预设匹配概率，概率最大的为最终检测结果，进而可以实时地进行在线检测。
四、反馈优化模块
未知位置的用户数据的检测结果，如果被判别为新的波束域，则对数据进行收集并进行标签处理，将其加入到训练集中，用于迭代训练，优化模型。
需要说明的是，本申请中使用的神经网络模型可以是卷积神经网络或全链接神经网络，但不限于此，可以是神经网络的任何一种。
需要说明的是，本申请实施例中基于聚类的波束域划分方法，主要包括：数据收集，数据预处理，聚类，波束域划分结果，图5是根据本公开实施例的基于聚类的波束域划分方法的流程图，如图5所示。需要说明的是，本申请实施例可以使用各种不同的聚类算法。
需要说明的是，本申请中使用机器学习进行波束域划分，可以是监督学习，也可以是无监督学习，不限于实施例中的基于神经网络的学习和基于聚类的学习。
为了使本申请的目的、技术方案更加清楚，下面结合具体实施例对本申请做进行详细描述。
实施例1
以NR上行SRS信道为例。
假设NR上行SRS占用32个资源块(Resource Block，简称为RB)，2RB绑定，用户端4端口，接收基站天线数为64，正负45度极化，垂直向阵子数N＝4，水平向阵子数M＝8。多用户位置位于(x＝100，y＝[-100 100]，z＝0)范围内。图6是根据本公开实施例的构建的多用户场景示意图(一)，(用户在y--100和y-100之间随机分布)。
本实施例进行波束域划分方法采用图3所示的四个阶段：数据初始收集与预处理、模型训练模块、在线检测模块、反馈优化模块。
一、数据初始收集与预处理
数据初始收集与预处理模块包括：数据收集、筛选处理、自动标签处理、数据预处理模 块。
a.数据收集
在初始阶段，收集一段时间的用户数据，本实施例中收集的是基站测量的SRS信道响应和SINR值。
b.筛选处理
本模块的目的主要是去除掉一些异常数据，使得训练样本数据尽可能的是真实有效的用户业务数据。本实施例中，将SINR低于-10dB的数据进行滤除。
c.自动标签处理
根据实际基站的覆盖范围，预置波束域栅格表。一般基站是三扇区覆盖，水平辐射范围为正负60度，下倾角一般在正负15度范围，按照水平和垂直5度的步长进行划分。则二维的预置波束域划分为如图4所示。根据估计出的垂直和水平向角度，判断归属预置波束域栅格的哪个栅格，进行标签的设置，这样训练用户数据与标签就建立起一一对应关系。
本实施例中，根据SRS信道估计值及天线阵列模式，计算垂直向协方差矩阵R v和水平向的协方差矩阵R h。将R v变换到角度域后，在角度域搜索峰值，根据峰值位置计算垂直角度θ。将R h变换到角度域后，在角度域搜索峰值，根据峰值位置计算水平角度
Figure PCTCN2022124391-appb-000001
根据估计的用户的垂直和水平角度
Figure PCTCN2022124391-appb-000002
判断其落在哪个栅格的角度范围内，结合预置波束域栅格表，进行标签的设置，这样训练用户数据与标签就建立起一一对应关系。本实施例中，利用的预置波束域栅格里的标签，即图4所示的有底色的数字标号。
d.数据预处理
提取多维数据特征，将数据预处理为机器学习所需要的数据格式。本实施例中，采用的SRS信道协方差矩阵的第一列。
SRS信道协方差矩阵Rx＝E[H*H H]，其中，H表示SRS信道响应，() H表示共轭转置。则神经网络或聚类算法的输入数据格式为Rx的第一列向量。由于目前应用的机器学习要求输入是实数，且输入值需要归一化，因此构造的输入特征数据格式为：
[real(normRx)；imag(normRx)] H
其中，Rx1＝Rx(：，1)；k mod＝max(|Rx1|)；normRx＝Rx1/k mod；输入特征数据的行数为样本数，列数为特征数。
二、模型训练模块
基于神经网络的学习训练模块主要分两个阶段：初始训练阶段、迭代训练阶段。
初始训练阶段：将初始预处理好的多维特征集数据分为两部分：训练集和测试集。把训练集输入到神经网络中进行训练，得到合适的模型参数，然后利用测试集进行模型的测试，得到测试结果；不断调整和优化模型参数，一直达到比较满意的测试结果为止，保存训练好的模型参数。
迭代训练阶段：如果未知位置用户的信道数据在线检测出来的归属各个波束域的匹配概率都低于某一个阈值，比如β＝0.35，那么会被判为新的波束域，保存当前用户的信道数据，进行模型的重新训练和优化。以此类推，这样不断扩大数据样本集，能够使网络模型不断优 化，检测出更多的波束域。
三、在线检测模块
将未知位置的用户的信道数据，进行同样的数据预处理，获取多维特征集，输入到训练好的神经网络中进行在线检测，给出预设匹配概率，概率最大的为最终检测结果。本实施例中，基于神经网络的空间波束域划分预测准确率达95％以上。
四、反馈优化模块
根据未知位置的用户数据的检测结果，如果被判别为新的波束域，则对数据进行收集并进行标签处理，将其加入到训练集中，用于迭代训练，优化模型。
本实施例采用神经网络模型为全链接四层神经网络，一个输入层，两个隐层，一个输出层，采用的优化器是随机梯度下降(Stochastic Gradient Descent，简称为SGD)，激活函数是Logistic。图7是根据本公开实施例中采用的全链接神经网络示意图。
实施例2
以NR上行SRS信道为例。
假设NR上行SRS占用104个RB，2RB绑定，用户端4端口，接收基站天线数为64，正负45度极化，垂直向阵子数N＝4，水平向阵子数M＝8。
图8是根据本公开实施例中构建的多用户场景示意图(二)。基于聚类算法的输入数据特征集也可以有多种，本实施例中采用的是信道协方差矩阵的第一列。
SRS信道协方差矩阵Rx＝E[H*H H]，其中，H表示SRS信道响应，() H表示共轭转置。由于目前应用的机器学习要求输入是实数，且输入值需要归一化，因此构造的输入特征数据格式为：
[real(normRx)；imag(normRx)] H
其中，Rx1＝Rx(：，1)；k mod＝max(|Rx1|)；normRx＝Rx1/k mod；输入特征数据的行数为样本数，列数为特征数。
本实施例中采用的是sklearn.cluster.KMeans及Bio.Cluste.kcluster模块来实现聚类算法，从而实现不同用户的波束域的划分。图9是根据本公开实施例中基于聚类的波束域划分结果示意图(横轴为样本数，纵轴为标签号，灰度深浅为聚类标识，同一灰度代表同一类，划分准确率为100％)。
另外，聚类方法和基于神经网络的方法可以联合使用，在聚类的先验结果上，进一步利用神经网络进行更精准的波束域划分。
与现有技术相比，本公开的有益效果如下：(1)在无线通信系统中的MU-MIMO技术中，基于机器学习进行空间波束域划分，更贴合实际的多变复杂的环境，有助提高配对成功率。(2)在线检测用户归属波束域，根据检测结果迭代训练检测，不断扩大训练数据集，易于实现，且可提高检测性能。(3)基于空间波束域的划分，空间特性统计，在应用MU-MIMO技术中，有助于降低用户配对复杂度，扩大用户配对选择范围，易选择到最优组合，获得最佳的配对增益。(4)通过对空间波束域的划分，在应用MU-MIMO技术中，有助于降低用户调度复 杂度。(5)本公开提取的多维特征集，易于计算，便于硬件实现。
此外，本申请实施例上述技术方案，可以进行空间波束域划分，对用户进行归属波束域检测，在MU-MIMO技术中，可以应用于多用户配对，多用户调度，提高MU-MIMO传输性能。在SU传输中，有助于最优波束的选择。在定位系统中，可以缩小定位区域，降低定位复杂度。在探测系统或监测系统中，在对目标的跟踪探测中，空间栅格化有助于降低复杂度，提高检测性能。
需要说明的是，本申请实施例在MU-MIMO场景，可以基于机器学习的用户选择、用户调度、多用户扎堆判断；在SU-MIMO场景，会涉及到空间最优波束域选择；在其他场景，比如定位系统中，探测系统中，监测系统中，也会涉及到空间域划分。
通过以上的实施方式的描述，本领域的技术人员可以清楚地了解到根据上述实施例的方法可借助软件加必需的通用硬件平台的方式来实现，当然也可以通过硬件，但很多情况下前者是更佳的实施方式。基于这样的理解，本公开的技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质(如ROM/RAM、磁碟、光盘)中，包括若干指令用以使得一台终端设备(可以是手机，计算机，服务器，或者网络设备等)执行本公开各个实施例的方法。
在本实施例中还提供了一种波束域的划分装置，该装置用于实现上述实施例及优选实施方式，已经进行过说明的不再赘述。如以下所使用的，术语“模块”可以实现预定功能的软件和/或硬件的组合。尽管以下实施例所描述的设备较佳地以软件来实现，但是硬件，或者软件和硬件的组合的实现也是可能并被构想的。
图10是根据本公开实施例的波束域的划分装置的结构框图，该装置包括：
获取模块10，设置为获取第一组测量数据；
第一划分模块12，设置为为所述第一组测量数据中的每一测量数据分配第一标签，并通过所述第一标签为所述每一测量数据在空间域中划分所述每一测量数据对应的波束域，其中，所述第一组测量数据中的不同测量数据对应不同的第一标签，所述波束域用于指示所述空间域在进行网格化的过程中划分的栅格；
训练模块14，设置为将具有所述第一标签的第一组测量数据输入神经网络模型中进行模型训练，得到波束域划分模型；
第二划分模块16，设置为通过所述波束域划分模型确定第二组测量数据中的每一测量数据对应的第二标签，以根据所述第二标签确定所述第二组测量数据中的每一测量数据在所述空间域中的目标波束域。
通过上述模块，获取第一组测量数据，并为第一组测量数据中的每一测量数据分配第一标签，进而通过第一标签为每一测量数据在空间域中划分每一测量数据对应的波束域，随后将具有第一标签的第一组测量数据输入神经网络模型中进行模型训练，得到波束域划分模型；通过波束域划分模型确定第二组测量数据中的每一测量数据对应的第二标签，以根据第二标签确定第二组测量数据中的每一测量数据在空间域中的目标波束域。采用上述技术方案，解决了传统方法采用预置波束对空间域进行栅格化操作，导致划分精度较低的问题。进而采用上述技术方案，利用神经网络进行波束域栅格划分，能够学习更复杂的更真实的空间域，使得栅格划分更准确可靠。
可选的，第二划分模块16还设置为在所述波束域划分模型中确定第一测量数据与第二测 量数据的相似度大于预设阈值的情况下，将所述第一测量数据的第一标签确定为所述第二测量数据的第二标签，其中，所述第一组测量数据包括：第一测量数据，所述第二组测量数据包括：第二测量数据；在所述波束域划分模型中确定第二测量数据与所述第一组测量数据中的每一测量数据的相似度都小于所述预设阈值的情况下，为所述第二测量数据分配第二标签，所述第二标签与所述第一组测量数据的所述第一标签不同。
可选的，第二划分模块16还设置为将所述第二标签中当前待划分波束域的第二标签与多个所述第一标签一一进行比对；在多个所述第一标签中存在与所述当前待划分波束域的第二标签相同的第一标签的情况下，获取与所述当前待划分波束域的第二标签相同的第一标签所对应的第一测量数据，并将所述当前待划分波束域的第二标签对应的测量数据分配至所述第一测量数据的波束域中；在多个所述第一标签中不存在与所述当前待划分波束域的第二标签相同的第一标签的情况下，根据所述当前待划分波束域的第二标签为所述当前待划分波束域的第二标签对应的测量数据在所述空间域中划分目标波束域。其中，所述第一标签的数量为多个，且所述第一标签的数量与所述第一组测量数据中测量数据的数量相同。
可选的，训练模块14还设置为将所述当前待划分波束域的第二标签对应的测量数据添加到所述波束域划分模型进行模型训练，得到目标波束域划分模型；将所述波束域划分模型替换为所述目标波束域划分模型。
可选的，训练模块14还设置为对具有所述第一标签的第一组测量数据中的每一测量数据进行特征提取，以得到所述每一测量数据的多维特征数据；将所述每一测量数据的多维特征数据的格式转化为目标格式，其中，所述目标格式为所述神经网络模型所匹配的格式；将具有所述目标格式的所述每一测量数据的多维特征数据输入所述神经网络模型中进行模型训练，得到波束域划分模型。
可选的，训练模块14还设置为将具有所述第一标签的第一组测量数据划分为训练集数据和测试集数据；将所述训练集数据输入所述神经网络模型中进行模型训练，得到第一神经网络模型；将所述测试集数据输入所述第一神经网络模型中，得到第一测试结果；在所述第一测试结果满足预设条件的情况下，将所述第一神经网络模型确定为波束域划分模型；在所述第一测试结果不满足预设条件的情况下，通过调整所述神经网络模型的参数配置再次进行模型训练。
可选的，训练模块14还设置为调整所述神经网络模型的参数配置，并将所述训练集数据添加到调整后的神经网络模型中进行训练，得到第二神经网络模型；将所述测试集数据输入所述第二神经网络模型中，得到第二测试结果；在所述第二测试结果满足预设条件的情况下，将所述第二神经网络模型确定为波束域划分模型。
可选的，第一划分模块12还设置为对第三组测量数据进行数据预处理，得到数据预处理后的第三组测量数据，其中，所述第三组测量数据包括：所述第一组测量数据，所述第二组测量数据；将所述数据预处理后的第三组测量数据通过聚类算法进行聚类处理，得到第三标签；通过所述第三标签为所述第三组测量数据中的每一测量数据在空间域中划分所述每一测量数据对应的波束域。
本公开的实施例还提供了一种计算机可读存储介质，该计算机可读存储介质中存储有计算机程序，其中，该计算机程序被设置为运行时执行上述任一项方法实施例中的步骤。
可选地，在本实施例中，上述存储介质可以被设置为存储用于执行以下步骤的计算机程 序：
S1，获取第一组测量数据；
S2，为所述第一组测量数据中的每一测量数据分配第一标签，并通过所述第一标签为所述每一测量数据在空间域中划分所述每一测量数据对应的波束域，其中，所述第一组测量数据中的不同测量数据对应不同的第一标签，所述波束域用于指示所述空间域在进行网格化的过程中划分的栅格；
S3，将具有所述第一标签的第一组测量数据输入神经网络模型中进行模型训练，得到波束域划分模型；
S4，通过所述波束域划分模型确定第二组测量数据中的每一测量数据对应的第二标签，以根据所述第二标签确定所述第二组测量数据中的每一测量数据在所述空间域中的目标波束域。
在一个示例性实施例中，上述计算机可读存储介质可以包括但不限于：U盘、只读存储器(Read-Only Memory，简称为ROM)、随机存取存储器(Random Access Memory，简称为RAM)、移动硬盘、磁碟或者光盘等各种可以存储计算机程序的介质。
本实施例中的具体示例可以参考上述实施例及示例性实施方式中所描述的示例，本实施例在此不再赘述。
本公开的实施例还提供了一种电子装置，包括存储器和处理器，该存储器中存储有计算机程序，该处理器被设置为运行计算机程序以执行上述任一项方法实施例中的步骤。
可选地，在本实施例中，上述处理器可以被设置为通过计算机程序执行以下步骤：
S1，获取第一组测量数据；
S2，为所述第一组测量数据中的每一测量数据分配第一标签，并通过所述第一标签为所述每一测量数据在空间域中划分所述每一测量数据对应的波束域，其中，所述第一组测量数据中的不同测量数据对应不同的第一标签，所述波束域用于指示所述空间域在进行网格化的过程中划分的栅格；
S3，将具有所述第一标签的第一组测量数据输入神经网络模型中进行模型训练，得到波束域划分模型；
S4，通过所述波束域划分模型确定第二组测量数据中的每一测量数据对应的第二标签，以根据所述第二标签确定所述第二组测量数据中的每一测量数据在所述空间域中的目标波束域。
在一个示例性实施例中，上述电子装置还可以包括传输设备以及输入输出设备，其中，该传输设备和上述处理器连接，该输入输出设备和上述处理器连接。
本实施例中的具体示例可以参考上述实施例及示例性实施方式中所描述的示例，本实施例在此不再赘述。
显然，本领域的技术人员应该明白，上述的本公开的各模块或各步骤可以用通用的计算装置来实现，它们可以集中在单个的计算装置上，或者分布在多个计算装置所组成的网络上，它们可以用计算装置可执行的程序代码来实现，从而，可以将它们存储在存储装置中由计算装置来执行，并且在某些情况下，可以以不同于此处的顺序执行所示出或描述的步骤，或者将它们分别制作成各个集成电路模块，或者将它们中的多个模块或步骤制作成单个集成电路模块来实现。这样，本公开不限制于任何特定的硬件和软件结合。
以上所述仅为本公开的优选实施例而已，并不用于限制本公开，对于本领域的技术人员来说，本公开可以有各种更改和变化。凡在本公开的原则之内，所作的任何修改、等同替换、改进等，均应包含在本公开的保护范围之内。

Claims (11)

  1. 一种波束域的划分方法，包括：
    获取第一组测量数据；
    为所述第一组测量数据中的每一测量数据分配第一标签，并通过所述第一标签为所述每一测量数据在空间域中划分所述每一测量数据对应的波束域，其中，所述第一组测量数据中的不同测量数据对应不同的第一标签，所述波束域用于指示所述空间域在进行网格化的过程中划分的栅格；
    将具有所述第一标签的第一组测量数据输入神经网络模型中进行模型训练，得到波束域划分模型；
    通过所述波束域划分模型确定第二组测量数据中的每一测量数据对应的第二标签，以根据所述第二标签确定所述第二组测量数据中的每一测量数据在所述空间域中的目标波束域。
  2. 根据权利要求1所述的波束域的划分方法，其中，通过所述波束域划分模型确定第二组测量数据中的每一测量数据对应的第二标签，包括：
    在所述波束域划分模型中确定第一测量数据与第二测量数据的相似度大于预设阈值的情况下，将所述第一测量数据的第一标签确定为所述第二测量数据的第二标签，其中，所述第一组测量数据包括：第一测量数据，所述第二组测量数据包括：第二测量数据；
    在所述波束域划分模型中确定第二测量数据与所述第一组测量数据中的每一测量数据的相似度都小于所述预设阈值的情况下，为所述第二测量数据分配第二标签，其中，所述第二标签与所述第一组测量数据的所述第一标签不同。
  3. 根据权利要求1所述的波束域的划分方法，其中，根据所述第二标签确定所述第二组测量数据中的每一测量数据在所述空间域中的目标波束域，包括：
    将所述第二标签中当前待划分波束域的第二标签与多个所述第一标签一一进行比对；
    在多个所述第一标签中存在与所述当前待划分波束域的第二标签相同的第一标签的情况下，获取与所述当前待划分波束域的第二标签相同的第一标签所对应的第一测量数据，并将所述当前待划分波束域的第二标签对应的测量数据分配至所述第一测量数据的波束域中；
    在多个所述第一标签中不存在与所述当前待划分波束域的第二标签相同的第一标签的情况下，根据所述当前待划分波束域的第二标签为所述当前待划分波束域的第二标签对应的测量数据在所述空间域中划分目标波束域。
  4. 根据权利要求3所述的波束域的划分方法，其中，根据所述当前待划分波束域的第二标签为所述当前待划分波束域的第二标签对应的测量数据在所述空间域中划分目标波束域之后，所述方法还包括：
    将所述当前待划分波束域的第二标签对应的测量数据添加到所述波束域划分模型进行模型训练，得到目标波束域划分模型；
    将所述波束域划分模型替换为所述目标波束域划分模型。
  5. 根据权利要求1所述的波束域的划分方法，其中，将具有所述第一标签的第一组测量数据输入神经网络模型中进行模型训练，得到波束域划分模型，包括：
    对具有所述第一标签的第一组测量数据中的每一测量数据进行特征提取，以得到所述每一测量数据的多维特征数据；
    将所述每一测量数据的多维特征数据的格式转化为目标格式，其中，所述目标格式为所述神经网络模型所匹配的格式；
    将具有所述目标格式的所述每一测量数据的多维特征数据输入所述神经网络模型中进行模型训练，得到波束域划分模型。
  6. 根据权利要求1所述的波束域的划分方法，其中，将具有所述第一标签的第一组测量数据输入神经网络模型中进行模型训练，得到波束域划分模型，包括：
    将具有所述第一标签的第一组测量数据划分为训练集数据和测试集数据；
    将所述训练集数据输入所述神经网络模型中进行模型训练，得到第一神经网络模型；
    将所述测试集数据输入所述第一神经网络模型中，得到第一测试结果；
    在所述第一测试结果满足预设条件的情况下，将所述第一神经网络模型确定为波束域划分模型；
    在所述第一测试结果不满足预设条件的情况下，通过调整所述神经网络模型的参数配置再次进行模型训练。
  7. 根据权利要求6所述的波束域的划分方法，其中，通过调整所述神经网络模型的参数配置再次进行模型训练，包括：
    调整所述神经网络模型的参数配置，并将所述训练集数据添加到调整后的神经网络模型中进行训练，得到第二神经网络模型；
    将所述测试集数据输入所述第二神经网络模型中，得到第二测试结果；
    在所述第二测试结果满足预设条件的情况下，将所述第二神经网络模型确定为波束域划分模型。
  8. 根据权利要求1所述的波束域的划分方法，其中，所述方法还包括：
    对第三组测量数据进行数据预处理，得到数据预处理后的第三组测量数据，其中，所述第三组测量数据包括：所述第一组测量数据，所述第二组测量数据；
    将所述数据预处理后的第三组测量数据通过聚类算法进行聚类处理，得到第三标签；
    通过所述第三标签为所述第三组测量数据中的每一测量数据在空间域中划分所述每一测量数据对应的波束域。
  9. 一种波束域的划分装置，包括：
    获取模块，设置为获取第一组测量数据；
    第一划分模块，设置为为所述第一组测量数据中的每一测量数据分配第一标签，并通过所述第一标签为所述每一测量数据在空间域中划分所述每一测量数据对应的波束域，其中，所述第一组测量数据中的不同测量数据对应不同的第一标签，所述波束域用于指示所述空间域在进行网格化的过程中划分的栅格；
    训练模块，设置为将具有所述第一标签的第一组测量数据输入神经网络模型中进行模型训练，得到波束域划分模型；
    第二划分模块，设置为通过所述波束域划分模型确定第二组测量数据中的每一测量数据对应的第二标签，以根据所述第二标签确定所述第二组测量数据中的每一测量数据在所述空间域中的目标波束域。
  10. 一种计算机可读的存储介质，所述存储介质中存储有计算机程序，其中，所述计算机程序被设置为运行时执行所述权利要求1至8任一项中所述的方法。
  11. 一种电子装置，包括存储器和处理器，所述存储器中存储有计算机程序，所述处理器被设置为通过所述计算机程序执行所述权利要求1至8任一项中所述的方法。
PCT/CN2022/124391 2021-10-29 2022-10-10 波束域的划分方法及装置、存储介质及电子装置 WO2023071760A1 (zh)

Applications Claiming Priority (4)

Application Number Priority Date Filing Date Title
CN202111277675.7 2021-10-29
CN202111277675 2021-10-29
CN202111511099.8 2021-12-10
CN202111511099.8A CN116090326A (zh) 2021-10-29 2021-12-10 波束域的划分方法及装置、存储介质及电子装置

Publications (1)

Publication Number Publication Date
WO2023071760A1 true WO2023071760A1 (zh) 2023-05-04

Family

ID=86159111

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/124391 WO2023071760A1 (zh) 2021-10-29 2022-10-10 波束域的划分方法及装置、存储介质及电子装置

Country Status (1)

Country Link
WO (1) WO2023071760A1 (zh)

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2018060812A1 (en) * 2016-09-30 2018-04-05 Nokia Technologies Oy Transceiver architecture for massive multiple-input multiple-output systems
CN108260133A (zh) * 2016-12-28 2018-07-06 维沃移动通信有限公司 一种波束测量上报的方法、网络侧设备及移动终端
US20190372644A1 (en) * 2018-06-01 2019-12-05 Samsung Electronics Co., Ltd. Method and apparatus for machine learning based wide beam optimization in cellular network
WO2020181533A1 (en) * 2019-03-13 2020-09-17 Nokia Shanghai Bell Co., Ltd. Device, method and computer readable medium for adjusting beamforming profiles
US20210126684A1 (en) * 2019-10-25 2021-04-29 Samsung Electronics Co., Ltd. Method and apparatus for antenna parameter configuration for cellular communication systems
CN113411110A (zh) * 2021-06-04 2021-09-17 东南大学 一种基于深度强化学习的毫米波通信波束训练方法

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2018060812A1 (en) * 2016-09-30 2018-04-05 Nokia Technologies Oy Transceiver architecture for massive multiple-input multiple-output systems
CN108260133A (zh) * 2016-12-28 2018-07-06 维沃移动通信有限公司 一种波束测量上报的方法、网络侧设备及移动终端
US20190372644A1 (en) * 2018-06-01 2019-12-05 Samsung Electronics Co., Ltd. Method and apparatus for machine learning based wide beam optimization in cellular network
WO2020181533A1 (en) * 2019-03-13 2020-09-17 Nokia Shanghai Bell Co., Ltd. Device, method and computer readable medium for adjusting beamforming profiles
US20210126684A1 (en) * 2019-10-25 2021-04-29 Samsung Electronics Co., Ltd. Method and apparatus for antenna parameter configuration for cellular communication systems
CN113411110A (zh) * 2021-06-04 2021-09-17 东南大学 一种基于深度强化学习的毫米波通信波束训练方法

Similar Documents

Publication Publication Date Title
KR100930799B1 (ko) 자동화된 클러스터링 방법 및 이를 이용한 이동통신환경에서 다중 경로의 클러스터링 방법 및 장치
CN109379153B (zh) 一种频谱感知方法
CN111869291A (zh) 无线系统中的信道制图
CN112040397B (zh) 一种基于自适应卡尔曼滤波的csi室内指纹定位方法
CN109299698B (zh) 一种基于支持向量机的无线信道场景识别方法
US8767671B2 (en) Space division multiple access transmission method of statistical characteristic mode
CN110324170A (zh) 数据分析设备、多模型共决策系统及方法
CN108667502A (zh) 一种基于机器学习的空间调制天线选择方法
CN110062410B (zh) 一种基于自适应谐振理论的小区中断检测定位方法
CN115486035A (zh) 用于信道估计的nn参数的类别
CN113965233A (zh) 一种基于深度学习的多用户宽带毫米波通信资源分配方法及系统
CN114844545A (zh) 基于sub6GHz信道以及部分毫米波导频的通信波束选择方法
Cheng et al. Channel path identification in mmWave systems with large-scale antenna arrays
Xiang et al. Robust sub-meter level indoor localization-a logistic regression approach
Liu et al. Location-based visible region recognition in extra-large massive MIMO systems
Ponnada et al. Best beam prediction in non-standalone mm wave systems
WO2023071760A1 (zh) 波束域的划分方法及装置、存储介质及电子装置
Su et al. Channel estimation in massive MIMO systems using a modified Bayes-GMM method
CN113273106A (zh) 上行链路数据流的处理
CN114269014B (zh) 基于域自适应网络的大规模mimo动态环境指纹定位方法
CN113746510B (zh) 基于机器学习的用户调度和模拟波束选择优化方法
Sun et al. Deep convolutional neural networks enabled fingerprint localization for massive MIMO-OFDM system
CN112996106B (zh) 一种去蜂窝大规模mimo系统定位方法
Shahabodini et al. Recurrent neural network and federated learning based channel estimation approach in mmWave massive MIMO systems
CN116090326A (zh) 波束域的划分方法及装置、存储介质及电子装置

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22885640

Country of ref document: EP

Kind code of ref document: A1