Unsupervised learning-based departure angle estimation method and device, and related medium

Abstract

The embodiment of the invention provides a departure angle estimation method and device based on unsupervised learning and a related medium, and belongs to the technical field of wireless communication. The method comprises the following steps: acquiring observation sequence data transmitted by a base station; pilot frequency information is obtained based on a communication protocol, and pilot frequency related information is obtained; when the pilot frequency related information is responded to be specific pilot frequency information, performing departure angle estimation on the observation sequence data and the pilot frequency information through a pre-trained first departure angle estimation model to obtain target departure angle information; and when the pilot frequency related information is responded to be pilot frequency statistical information of the pilot frequency information, performing departure angle estimation on the observation sequence data and the pilot frequency statistical information through a pre-trained second departure angle estimation model to obtain target departure angle information. According to the embodiment of the invention, the terminal equipment can accurately estimate the departure angle information of the signal transmitted by the base station terminal.

Description

Technical Field

[0001] The present application relates to the field of wireless communication technology, and in particular to a departure angle estimation method, device and related medium based on unsupervised learning. Background Art

[0002] Angle estimation plays a vital role in many fields such as astronomy, navigation, radar sensing, etc. The downlink positioning method refers to sending signals from the base station to the terminal device, which receives and processes these signals to determine its own position. The Angle of Departure (AOD) refers to the angle at which the signal leaves the transmitting antenna array.

[0003] In the related technologies, many IoT devices need to realize the positioning of the devices through downlink angle estimation. However, most devices in the IoT are only equipped with a single antenna. Many related angle estimation methods are mainly designed for multi-antenna array systems, which are composed of multiple antennas. These angle estimation methods are not suitable for IoT devices with only a single antenna. The antenna resources of IoT devices are scarce, which makes it very difficult for single-antenna devices to estimate the departure angle of the transmitter and difficult to obtain accurate results. Therefore, how to enable single-antenna terminal devices to accurately estimate the departure angle information of the base station terminal's transmitted signal has become a technical problem that needs to be solved urgently. Summary of the invention

[0004] The main purpose of the embodiments of the present application is to propose a departure angle estimation method, device and related medium based on unsupervised learning, aiming to enable a terminal device to accurately estimate the departure angle information of a signal transmitted by a base station terminal.

[0005] To achieve the above object, a first aspect of an embodiment of the present application proposes a departure angle estimation method based on unsupervised learning, the method comprising:

[0006] Obtaining observation sequence data transmitted from a base station;

[0007] Acquire pilot information based on the communication protocol to obtain pilot related information;

[0008] In response to the pilot-related information being specific pilot information, performing departure angle estimation on the observation sequence data and the pilot information by using a pre-trained first departure angle estimation model to obtain target departure angle information;

[0009] In response to the pilot-related information being pilot statistical information of the pilot information, a pre-trained second departure angle estimation model is used to perform departure angle estimation on the observation sequence data and the pilot statistical information to obtain target departure angle information.

[0010] In some embodiments, the observation sequence data includes a first number of observations;

[0011] The step of estimating the departure angle of the observation sequence data and the pilot information by using a pre-trained first departure angle estimation model to obtain target departure angle information includes:

[0012] Based on the pilot information, construct a pilot signal matrix;

[0013] Generate a first model input tensor according to the pilot signal matrix and a first number of the observation values;

[0014] Inputting the first model input tensor into the first pre-trained departure angle estimation model to perform departure angle estimation to obtain target departure angle information;

[0015] The step of estimating the angle of departure of the observation sequence data and the pilot statistical information by using a pre-trained second angle of departure estimation model to obtain target angle of departure information includes:

[0016] Based on the pilot statistical information, construct a pilot vector matrix;

[0017] Perform Kronecker product processing based on the pilot vector matrix to obtain a pilot processing matrix;

[0018] generating a second model input tensor according to the pilot processing matrix and the first number of the observations;

[0019] The second model input tensor is input into the pre-trained second departure angle estimation model to perform departure angle estimation to obtain target departure angle information.

[0020] In some embodiments, before performing departure angle estimation on the observation sequence data and the pilot information by using a pre-trained first departure angle estimation model to obtain target departure angle information, the first departure angle estimation model is also pre-trained, including:

[0021] Acquire a first training data set; the first training data set includes first observation training data and pilot training data;

[0022] Inputting the first observed training data and the pilot training data into an original first departure angle estimation model, and outputting first predicted departure angle information and a first channel gain estimation value;

[0023] Based on deterministic maximum likelihood estimation, unsupervised learning is performed on the original first angle of departure estimation model according to the first predicted angle of departure information and the first channel gain estimation value to obtain the pre-trained first angle of departure estimation model.

[0024] In some embodiments, the performing unsupervised learning on the original first angle of departure estimation model based on the deterministic maximum likelihood estimation according to the first predicted angle of departure information and the first channel gain estimation value to obtain the pre-trained first angle of departure estimation model includes:

[0025] Calculate a steering vector based on the first predicted departure angle information to obtain a first predicted angle steering vector;

[0026] Constructing a first model loss function based on the pilot training data, the first prediction angle steering vector, the first channel gain estimate and the first observation training data:

[0027]

[0028] Wherein, X1 represents the pilot training data, Y1 represents the first observed training data, D1 represents the first training data set, W1 represents the first model parameter of the original first departure angle estimation model, ξ1 represents the first channel gain estimation value, θ1 represents the first predicted departure angle information, and a(θ1) represents the first predicted angle steering vector;

[0029] The first model parameters of the original first departure angle estimation model are adjusted based on the first model loss function to obtain the pre-trained first departure angle estimation model.

[0030] In some embodiments, before performing departure angle estimation on the observation sequence data and the pilot statistical information by using a pre-trained second departure angle estimation model to obtain target departure angle information, the second departure angle estimation model is also pre-trained, including:

[0031] Acquire a second training data set; the second training data set includes second observed training data and pilot training statistical data;

[0032] Inputting the second observed training data and the pilot training statistical data into an original second departure angle estimation model, and outputting second predicted departure angle information, a second channel gain estimation value and noise variance information;

[0033] generating a reconstructed covariance matrix based on the second predicted departure angle information, the second channel gain estimate and the noise variance information, and generating a sample covariance matrix based on the second observed training data;

[0034] Based on stochastic maximum likelihood estimation, unsupervised learning is performed on the original second departure angle estimation model according to the reconstructed covariance matrix and the sample covariance matrix to obtain a pre-trained second departure angle estimation model.

[0035] In some embodiments, generating a reconstructed covariance matrix based on the second predicted departure angle information, the second channel gain estimate, and the noise variance information, and generating a sample covariance matrix based on the second observed training data includes:

[0036] Calculate a steering vector based on the second predicted departure angle information to obtain a second predicted angle steering vector;

[0037] Generate a reconstructed covariance matrix based on the second prediction angle steering vector, the second channel gain estimation value and the noise variance information:

[0038]

[0039] Among them, C y represents the reconstructed covariance matrix, ξ2 represents the second channel gain estimate, w (q) represents the noise variance information, θ2 represents the second predicted departure angle information, a(θ2) represents the second predicted angle steering vector, X (q) (X (q) ) H represents the covariance of the pilot information in the pilot training statistical data, Indicates the expected value.

[0040] Generate a sample covariance matrix based on the second observed training data:

[0041]

[0042] in, represents the sample covariance matrix, Q represents the number of the second observed training data, y (q) represents the qth data of the second observed training data, (y (q) ) H Represents y (q) The conjugate transposed matrix of .

[0043] In some embodiments, the stochastic maximum likelihood estimation is based on performing unsupervised learning on the original second departure angle estimation model according to the reconstructed covariance matrix and the sample covariance matrix to obtain the pre-trained second departure angle estimation model, specifically including:

[0044] Construct a second model loss function based on the reconstructed covariance matrix and the sample covariance matrix:

[0045]

[0046] Among them, C y represents the reconstructed covariance matrix, represents the sample covariance matrix, X2 represents the pilot training statistical data, Y2 represents the second observed training data, D2 represents the second training data set, W2 represents the second model parameter of the original second departure angle estimation model, tr() refers to the trace of the matrix, that is, the sum of the diagonal elements of the main diagonal of the matrix, det(C y ) represents the determinant value of the reconstructed covariance matrix;

[0047] The second model parameters of the original second departure angle estimation model are adjusted based on the second model loss function to obtain a pre-trained second departure angle estimation model.

[0048] To achieve the above object, a second aspect of an embodiment of the present application proposes a departure angle estimation device based on unsupervised learning, the device comprising:

[0049] An observation data acquisition module is used to obtain observation sequence data transmitted from the base station;

[0050] A pilot information acquisition module is used to acquire pilot information based on a communication protocol to obtain pilot related information;

[0051] A first angle estimation module is used for, in response to the pilot-related information being the specific pilot information, estimating the departure angle on the observation sequence data and the pilot information by using a pre-trained first departure angle estimation model to obtain target departure angle information;

[0052] The second angle estimation module is used to estimate the departure angle of the observation sequence data and the pilot statistical information by using a pre-trained second departure angle estimation model to obtain target departure angle information in response to the pilot related information being the pilot statistical information of the pilot information.

[0053] To achieve the above-mentioned purpose, the third aspect of an embodiment of the present application proposes an electronic device, which includes a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, it implements the departure angle estimation method based on unsupervised learning described in the first aspect above.

[0054] To achieve the above-mentioned purpose, the fourth aspect of an embodiment of the present application proposes a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the departure angle estimation method based on unsupervised learning described in the first aspect above.

[0055] The present application proposes a method, device and related medium for estimating an angle of departure based on unsupervised learning, which obtains pilot related information by acquiring observation sequence data transmitted by a base station and acquiring pilot information based on a communication protocol. In response to the pilot related information being specific pilot information, the observation sequence data and the pilot information are estimated by a pre-trained first angle of departure estimation model to obtain target angle of departure information; in response to the pilot related information being pilot statistical information of the pilot information, the observation sequence data and the pilot statistical information are estimated by a pre-trained second angle of departure estimation model to obtain target angle of departure information. Therefore, in order to improve the accuracy of the angle of departure estimation, the present application not only uses the received observation sequence data as the input of the neural network model, but also uses the pilot information as the input. According to the obtained pilot related information, if the pilot related information is specific pilot information, the first angle of departure estimation model is used to estimate the angle of departure. If the pilot related information is pilot statistical information, the second angle of departure estimation model is used to estimate the angle of departure, so that the user equipment can estimate the angle of departure according to different states of the pilot information and obtain accurate target angle of departure information. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 is a flow chart of a departure angle estimation method based on unsupervised learning provided in an embodiment of the present application;

[0057] Figure 2A yes Figure 1 Flow chart of step S103 in FIG.

[0058] Figure 2B yes Figure 1 Flow chart of step S104 in FIG.

[0059] Figure 3 yes Figure 1 The flowchart before step S103 in FIG.

[0060] Figure 4 yes Figure 3 Flow chart of step S303 in FIG.

[0061] Figure 5 yes Figure 3 The flowchart before step S104 in FIG.

[0062] Figure 6 yes Figure 5 Flow chart of step S503 in FIG.

[0063] Figure 7 yes Figure 5 Flow chart of step S504 in FIG.

[0064] Figure 8is a graph showing the relationship between the mean absolute error index of different methods provided in the embodiments of the present application and the sampling time;

[0065] Fig. 9 is a graph showing the relationship between the mean absolute error index of different methods provided in the embodiments of the present application and the base station transmission power;

[0066] Fig.10 is a structural schematic diagram of a departure angle estimation device based on unsupervised learning provided in an embodiment of the present application;

[0067] Fig.11 It is a schematic diagram of the hardware structure of the electronic device provided in the embodiment of the present application. DETAILED DESCRIPTION

[0068] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0069] It should be noted that, although the functional modules are divided in the device schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first", "second", etc. in the specification, claims and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0070] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0071] Artificial intelligence (AI) is a new technical science that studies and develops theories, methods, technologies and application systems for simulating, extending and expanding human intelligence. AI is a branch of computer science that attempts to understand the essence of intelligence and produce a new type of intelligent machine that can respond in a similar way to human intelligence. Research in this field includes robotics, speech recognition, image recognition, natural language processing and expert systems. AI can simulate the information process of human consciousness and thinking. AI is also a theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0072] Unsupervised learning is a branch of machine learning that deals with unlabeled data sets. Unlike supervised learning, in unsupervised learning, the algorithm is not told which data belongs to the same category or the value of the target variable that needs to be predicted, but needs to find the structure, pattern or relationship in the data by itself.

[0073] Downlink (DL) refers to the process of sending information from a base station (BS) or other network nodes to a terminal device (such as a mobile phone, computer or other user equipment UE).

[0074] Channel State Information (CSI), a term used in the field of wireless communications, refers to the channel properties of a communication link. It describes the attenuation factor of a signal on each transmission path, that is, the value of each element in the channel gain matrix H, such as signal scattering, environmental fading (fading, multipath fading or shadowing fading), and power decay of distance. CSI can adapt the communication system to the current channel conditions and provide a guarantee for high-reliability and high-speed communications in a multi-antenna system.

[0075] The embodiments of the present application provide a departure angle estimation method, apparatus and related media based on unsupervised learning, which are intended to enable a terminal device to accurately estimate the departure angle information of a signal transmitted by a base station terminal.

[0076] The departure angle estimation method, device and related medium based on unsupervised learning provided in the embodiments of the present application are specifically illustrated through the following embodiments. First, the departure angle estimation method based on unsupervised learning in the embodiments of the present application is described.

[0077] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0078] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0079] The departure angle estimation method based on unsupervised learning provided in the embodiment of the present application relates to the field of wireless communication technology. The departure angle estimation method based on unsupervised learning provided in the embodiment of the present application can be applied to a terminal, can also be applied to a server side, and can also be software running in a terminal or a server side. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, etc.; the server side can be configured as an independent physical server, or a server cluster or a distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application that implements the departure angle estimation method based on unsupervised learning, etc., but is not limited to the above forms.

[0080] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0081] Figure 1 is an optional flowchart of a departure angle estimation method based on unsupervised learning provided in an embodiment of the present application. Figure 1 The method may include but is not limited to steps S101 to S104.

[0082] Step S101, obtaining observation sequence data transmitted from a base station;

[0083] Step S102, obtaining pilot information based on the communication protocol to obtain pilot related information;

[0084] Step S103, in response to the pilot-related information being specific pilot information, performing departure angle estimation on the observation sequence data and the pilot information by using a pre-trained first departure angle estimation model to obtain target departure angle information;

[0085] Step S104, in response to the pilot-related information being pilot statistical information of the pilot information, performing departure angle estimation on the observation sequence data and the pilot statistical information by using a pre-trained second departure angle estimation model to obtain target departure angle information.

[0086] In the steps S101 to S106 shown in the embodiment of the present application, the observation sequence data transmitted by the base station is obtained, and the pilot information is obtained based on the communication protocol to obtain the pilot related information. In response to the pilot related information being specific pilot information, the observation sequence data and the pilot information are estimated by the first departure angle estimation model trained in advance to obtain the target departure angle information; in response to the pilot related information being the pilot statistical information of the pilot information, the observation sequence data and the pilot statistical information are estimated by the second departure angle estimation model trained in advance to obtain the target departure angle information. Therefore, in order to improve the accuracy of departure angle estimation, the present application not only uses the received observation sequence data as the input of the neural network model, but also uses the pilot information as the input. According to the obtained pilot related information, if the pilot related information is specific pilot information, the first departure angle estimation model is used to estimate the departure angle. If the pilot related information is pilot statistical information, the second departure angle estimation model is used to estimate the departure angle, so that the user equipment can estimate the departure angle according to different states of the pilot information and obtain accurate target departure angle information.

[0087] In step S101 of some embodiments, observation sequence data is received and acquired from a base station. In a wireless communication system, observation sequence data generally refers to actual measurement values ​​of signals periodically sent by a base station. These signals are affected by various factors during propagation to user equipment, such as path loss, multipath effect, signal attenuation, etc. These observation sequence data contain rich information and can be used to estimate the propagation characteristics of the signal, including the arrival angle or departure angle of the signal. User equipment refers to most devices in the Internet of Things.

[0088] In step S102 of some embodiments, pilot information is obtained through a communication protocol, but some communication protocols may not publish specific pilot information, but publish statistical information about the pilot information, i.e., pilot statistical information. Pilot statistical information here refers to first-order statistical information and second-order statistical information of the pilot information, i.e., expected value, variance, and covariance information. Therefore, when the pilot information is obtained through the communication protocol, the pilot-related information obtained may be specific pilot information or pilot statistical information.

[0089] For user equipment, pilot information can be obtained through communication protocols and is a known signal. The base station can periodically send this pilot information, and the user equipment estimates the channel state information (CSI) by comparing the received pilot information (that is, the observation sequence data) with the known pilot information. In the case of a single-antenna device, due to the lack of spatial diversity in a multi-antenna system, the acquisition and processing of pilot information becomes more important and is the main source of information for departure angle estimation.

[0090] After the pilot-related information is acquired, it is necessary to input the information into the first departure angle estimation model or the second departure angle estimation model to perform departure angle estimation, depending on whether it is specific pilot information or pilot statistical information.

[0091] In step S103 of some embodiments, if the pilot-related information obtained is specific pilot information, the observation sequence data and the pilot information are input into a pre-trained first angle of departure estimation model to perform angle of departure estimation to obtain target angle of departure information. The user equipment will use the pre-trained first angle of departure estimation model to process the observation sequence data and the pilot information. This model is designed based on the deterministic maximum likelihood (DML) estimation principle, which uses specific pilot information to estimate the angle of departure (AoD) of the transmitted signal.

[0092] See also Figure 2A In some embodiments, the observation sequence data includes a first number of observation values, and step S103 may include but is not limited to steps S201 to S203:

[0093] Step S201, constructing a pilot signal matrix based on pilot information;

[0094] Step S202, generating a first model input tensor according to the pilot signal matrix and a first number of observation values;

[0095] Step S203, inputting the first model input tensor into a pre-trained first departure angle estimation model to perform departure angle estimation to obtain target departure angle information;

[0096] See also Figure 2B In some embodiments, step S104 may include but is not limited to steps S204 to S206:

[0097] Step S204, constructing a pilot vector matrix based on the pilot statistical information;

[0098] Step S205, performing Kronecker product processing based on the pilot vector matrix to obtain a pilot processing matrix;

[0099] Step S206, generating a second model input tensor according to the pilot processing matrix and the first number of observations;

[0100] Step S207: input the second model input tensor into a pre-trained second departure angle estimation model to perform departure angle estimation to obtain target departure angle information.

[0101] In step S201 of some embodiments, the observation sequence data is the pilot information transmitted by the user equipment from the base station during a transmission period. The transmission duration T of the observation sequence data can be divided into a first number Q equal-length blocks, each block being L unit durations. Thus, T=Q*L. is the pilot information transmitted in block q,

[0102] Assuming that the user equipment remains stationary during Q blocks and receives the observation sequence data, the observation value in block q can be expressed as where Y (q) The value in can be calculated by formula (1), which is as follows:

[0103] Y (q) =ξX (q) a(θ)+w (q) (1)

[0104] Among them, w (q) It represents additive white Gaussian noise. Additive white Gaussian noise (AWGN) describes the random noise superimposed on the transmission signal during the signal transmission process of the wireless communication system. This noise model is very important in both theoretical and practical communication system design. It affects the signal quality and system performance. The amplitude distribution of additive white Gaussian noise follows the Gaussian distribution. ξ represents the channel gain parameter, which is determined by the distance between the base station and the user equipment. Since the user equipment remains stationary during Q blocks, the channel gain parameter can be considered to be a constant value. θ is the departure angle, and a(θ) represents the steering vector calculated based on the departure angle.

[0105] Therefore, the first number of observations of the observation sequence data can be expressed as Corresponding to the observed sequence data, the first number of pilot information is constructed into a pilot information matrix, and the

[0106] In steps S202 to S203 of some embodiments, the pilot signal matrix and the first number of observation values ​​are used to generate a first model input tensor, wherein the first number of observation values ​​are The first model input tensor S is obtained as:

[0107]

[0108] Among them, Y L,Q represents the (L, Q)th item in the first number of observations B, X M,L represents the (M, L)th item in the first number of pilot signals A. Then the first model input tensor S is divided into a real part and an imaginary part, and input into a pre-trained first departure angle estimation model to perform departure angle estimation to obtain target departure angle information. The target departure angle information includes the departure angle at which the base station sends a signal to the user equipment, which is estimated by the model.

[0109] In steps S204 to S205 of some embodiments, the pilot information The generation of where c (q) is the value of the pilot information transmitted in the qth block, is the beamforming vector, and the pilot statistics can be understood as c (q) The first-order and second-order statistical information of the beamforming vector is still known, and the pilot vector matrix is ​​obtained Then, the pilot processing matrix is ​​obtained by performing Kronecker product on the pilot vector matrix and the all-1 vector of the first dimension.

[0110] In steps S206 to S207 of some embodiments, the pilot processing matrix and the first number of observations are used to generate a second model input tensor, wherein the first number of observations is The second model input tensor S is obtained. The specific steps are the same as step S202, except that the pilot processing matrix C and the first number of observation values ​​B constitute the second model input tensor S, and then the second model input tensor is divided into a real part and an imaginary part, which are input into a pre-trained second departure angle estimation model to perform departure angle estimation to obtain the target departure angle information.

[0111] See also Figure 3 In some embodiments, before step S103, the first departure angle estimation model is pre-trained, which may include but is not limited to steps S301 to S303:

[0112] Step S301, obtaining a first training data set; the first training data set includes first observation training data and pilot training data;

[0113] Step S302, inputting the first observed training data and the pilot training data into the original first departure angle estimation model, and outputting first predicted departure angle information and a first channel gain estimation value;

[0114] Step S303 , based on deterministic maximum likelihood estimation, unsupervised learning is performed on the original first angle of departure estimation model according to the first predicted angle of departure information and the first channel gain estimation value to obtain a pre-trained first angle of departure estimation model.

[0115] In step S301 of some embodiments, a first training data set is obtained, which includes first observation training data for simulating real-world signal propagation conditions and specific pilot information for training, i.e., pilot training data. A first departure angle estimation model is pre-trained using the first observation training data and the pilot training data, so that the pre-trained first departure angle estimation model can accurately estimate the departure angle information for the specific pilot information.

[0116] In step S302 of some embodiments, the first observed training data and the pilot training data are input into the original first departure angle estimation model to perform angle estimation, and the obtained prediction result includes first predicted departure angle information and a first channel gain estimation value.

[0117] In step S303 of some embodiments, unsupervised learning is performed on the original first departure angle estimation model based on the prediction results obtained in step S302 and based on the deterministic maximum likelihood estimation theory. Unsupervised learning is a training method that does not rely on labeled data, which allows the model to improve its performance through self-learning and self-adjustment. At this stage, the model may use various optimization algorithms to adjust its internal parameters so that it can provide more accurate predictions when encountering similar input data in the future. In this way, the model can gradually adapt to the characteristics of the training data and eventually obtain a pre-trained first departure angle estimation model.

[0118] Through step S301 to step S303, the first departure angle estimation model is pre-trained so that it can provide accurate departure angle information based on specific pilot information.

[0119] See also Figure 4 In some embodiments, step S303 may include but is not limited to steps S401 to S403:

[0120] Step S401, calculating a steering vector based on first predicted departure angle information to obtain a first predicted angle steering vector;

[0121] Step S402, constructing a first model loss function based on pilot training data, a first prediction angle steering vector, a first channel gain estimation value and first observation training data;

[0122] Step S403: adjusting the first model parameters of the original first departure angle estimation model based on the first model loss function to obtain a pre-trained first departure angle estimation model.

[0123] In step S401 of some embodiments, before calculating the steering vector, the pilot training data is parsed or the signal wavelength, the number of base station antennas and the base station antenna spacing are obtained according to the communication protocol. The base station generally actively broadcasts the configuration information about the base station, such as the number of base station antennas.

[0124] Then, the steering vector is calculated based on the signal wavelength, the number of base station antennas, the base station antenna spacing and the first predicted departure angle information, and the following formula (2) is obtained:

[0125]

[0126] Among them, a(θ1) is the predicted angle steering vector, θ1 is the first predicted departure angle information, λ represents the signal wavelength, M represents the number of base station antennas equipped by the base station, d represents the base station antenna spacing between antennas configured by the base station, and j represents the imaginary part of the complex number.

[0127] In step S402 of some embodiments, a first model loss function is constructed based on pilot training data, a first prediction angle steering vector, a first channel gain estimate, and first observation training data.

[0128] If the user equipment fully knows the pilot information X of all blocks q∈[Q] (q) , that is, the specific pilot information, then the actual measurement value Y (q) In formula (1), it obeys the complex Gaussian distribution. Based on formula (1), the log-likelihood p(y (1) ,…,y (Q) ; The maximum likelihood estimation of θ1,ξ1) is equivalent to solving the least squares problem and obtaining the first model loss function, as shown in formula (3):

[0129]

[0130] Among them, X1 represents the pilot training data, Y1 represents the signal of the first observed training data, D1 represents the first training data set, W1 represents the first model parameter of the original first departure angle estimation model, ξ1 represents the first channel gain estimate, θ1 represents the first predicted departure angle information, and a(θ1) represents the first predicted angle steering vector.

[0131] In step S403 of some embodiments, the first model loss function is obtained through step S402, and the original first departure angle estimation model can learn how to better estimate the departure angle and channel gain according to the first observation training data and the pilot training data. With the continuous adjustment of parameters, the performance of the model will be improved, and finally a pre-trained first departure angle estimation model is obtained.

[0132] With unsupervised learning, models can be trained without real labeled data, which is extremely valuable in real-world applications where obtaining such labeled data can be difficult and time-consuming.

[0133] Through step S401 to step S403, the original first departure angle estimation model is pre-trained using the first model loss function, so that accurate departure angle estimation can be provided based on specific pilot information.

[0134] In step S104 of some embodiments, if the pilot-related information obtained is statistical information of the pilot information, the user equipment will use a pre-trained second departure angle estimation model. This model is designed based on the stochastic maximum likelihood (SML) estimation principle, and it can perform angle estimation when only the first-order statistical information and the second-order statistical information of the pilot information are available. Unlike the deterministic model, the stochastic model uses the statistical information of the pilot information, such as the covariance matrix of the pilot information, to estimate the departure angle of the signal.

[0135] See also Figure 5 In some embodiments, before step S104, the second departure angle estimation model is also pre-trained, which may include but is not limited to steps S501 to S504:

[0136] Step S501, obtaining a second training data set; the second training data set includes second observed training data and pilot training statistical data;

[0137] Step S502, inputting the second observed training data and the pilot training statistical data into the original second departure angle estimation model, and outputting second predicted departure angle information, second channel gain estimation value and noise variance information;

[0138] Step S503, generating a reconstructed covariance matrix based on the second predicted departure angle information, the second channel gain estimation value and the noise variance information, and generating a sample covariance matrix based on the second observed training data;

[0139] Step S504 , based on stochastic maximum likelihood estimation, unsupervised learning is performed on the original second departure angle estimation model according to the reconstructed covariance matrix and the sample covariance matrix to obtain a pre-trained second departure angle estimation model.

[0140] In step S501 of some embodiments, a second training data set is obtained, which includes second observed training data and pilot training statistics data for simulating real-world signal propagation conditions. The second observed training data represents a signal received from a base station, and the pilot training statistics data provides first-order statistics and second-order statistics of pilot information. Thus, the pre-trained second departure angle estimation model can accurately estimate departure angle information when only the statistics of the pilot information are known.

[0141] In step S502 of some embodiments, the second observed training data and the pilot training statistical data are input into the original second departure angle estimation model to perform angle estimation, and the obtained prediction results include second predicted departure angle information, second channel gain estimation value and noise variance information.

[0142] In step S503 of some embodiments, a reconstructed covariance matrix is ​​generated based on the prediction results obtained in step S502. This matrix reflects the estimation of the signal and noise characteristics by the second departure angle estimation model. At the same time, a sample covariance matrix is ​​generated based on the second observed training data, and this matrix represents the statistical information of the actual received signal, that is, the statistical information of the observed data. These two covariance matrices will be used to evaluate the performance of the model.

[0143] See also Figure 6 In some embodiments, step S503 may include but is not limited to steps S601 to S603:

[0144] Step S601, calculating a steering vector based on the second predicted departure angle information to obtain a second predicted angle steering vector;

[0145] Step S602, generating a reconstructed covariance matrix based on the second prediction angle steering vector, the second channel gain estimation value and the noise variance information;

[0146] Step S603: Generate a sample covariance matrix based on the second observed training data.

[0147] In step S601 of some embodiments, the operation here is the same as step S401, except that the parameter based on which the calculation is performed is the second predicted departure angle information.

[0148] In step S602 of some embodiments, due to the pilot training statistical data obtained by the user equipment, only statistical information about the pilot information, such as the mean, variance and covariance, can be obtained.

[0149] A reconstructed covariance matrix is ​​generated based on the second prediction angle steering vector, the second channel gain estimation value and the noise variance information, as shown in the following formula (4):

[0150]

[0151] Among them, C y represents the reconstructed covariance matrix, ξ2 represents the second channel gain estimate, and w (q) represents the noise variance information, θ2 represents the second predicted departure angle information, a(θ2) represents the second predicted angle steering vector, X (q) (X (q) ) H represents the covariance of the pilot information in the pilot training statistical data, Indicates the mean.

[0152] In step S603 of some embodiments, a sample covariance matrix is ​​generated based on the second observed training data, as follows:

[0153]

[0154] in, represents the sample covariance matrix, Q represents the number of second observed training data, y (q) represents the qth data of the second observed training data, (y (q) ) H Represents y (q) The conjugate transposed matrix of .

[0155] Through steps S601 to S603, necessary input and feedback are provided for the training of the second departure angle estimation model. By reconstructing the covariance matrix and the sample covariance matrix to construct a loss function, the prediction performance of the model can be evaluated and the model parameters can be adjusted accordingly. This method of constructing a loss function based on the covariance matrix allows the model to perform unsupervised learning without real labeled data, thereby optimizing its prediction ability.

[0156] In step S504 of some embodiments, a loss function is constructed based on the reconstructed covariance matrix and the sample covariance matrix, and unsupervised learning is performed on the original second departure angle estimation model. This process involves using the loss function to adjust the model parameters, in this way, the model can learn how to better estimate the departure angle based on the statistical information of the pilot information. As the parameters are continuously adjusted, the performance of the model will be improved, and finally a pre-trained second departure angle estimation model is obtained.

[0157] See also Figure 7 In some embodiments, step S504 may include but is not limited to steps S701 to S702:

[0158] Step S701, constructing a second model loss function based on the reconstructed covariance matrix and the sample covariance matrix;

[0159] Step S702: adjust the second model parameters of the original second departure angle estimation model based on the second model loss function to obtain a pre-trained second departure angle estimation model.

[0160] In step S701 of some embodiments, it can be known from the statistical information of the pilot training statistical data that the mean value of the pilot in the lth time slot of the qth block of all pilot training statistical data is l∈[L] and q∈[Q], and the correlation between the pilot information of the lth and l′th time slots Therefore, the second observed training data obeys the complex Gaussian distribution. Therefore, based on formula (4) and formula (5), we can get the log-likelihood p(y (1) ,…,y (Q) ; The maximum likelihood estimation problem of θ2,ξ2) is equivalent to the minimization problem, and the loss function of the second model is obtained as follows:

[0161]

[0162] Among them, C y represents the reconstructed covariance matrix, represents the sample covariance matrix, X2 represents the pilot training statistics, Y2 represents the second observed training data, D2 represents the second training data set, W2 represents the second model parameter of the original second departure angle estimation model, tr() refers to the trace of the matrix, that is, the sum of the diagonal elements of the main diagonal of the matrix, det(C y ) represents the determinant value of the reconstructed covariance matrix;

[0163] In step S702 of some embodiments, the second model loss function is obtained through step S402, and the original second departure angle estimation model can learn how to better estimate the departure angle and channel gain based on the first observation training data and the pilot training data. With the continuous adjustment of parameters, the performance of the model will be improved, and finally a pre-trained first departure angle estimation model is obtained.

[0164] Through step S701 to step S702, the original second departure angle estimation model is pre-trained using the second model loss function, so that accurate departure angle estimation can be provided according to the pilot statistical information.

[0165] Through steps S501 to S504, by constructing a reconstructed covariance matrix and a sample covariance matrix, unsupervised learning is performed on the original second departure angle estimation model based on the stochastic maximum likelihood estimation theory to obtain a pre-trained second departure angle estimation model, so that accurate target departure angle information can be obtained based on the observed sequence data and pilot statistical information.

[0166] In some experimental embodiments, the mean absolute error (MAE) indicator is used to evaluate the error performance of the estimated departure angle. Figure 8 , set up the same experimental environment, and use simulation or actual measurement data to build training sets and validation sets, test different sampling time slots L and different base station transmission power P, and various methods to obtain the mean absolute error (MAE) index of the estimated departure angle. In the figure, ESPRIT represents the method of estimating signal parameters via rotational invariance techniques (ESPRIT), MUSIC represents the multiple signal classification algorithm (MUSIC), AE-based represents the signal processing method based on autoencoder (Autoencoder), DFT-based represents the signal processing method based on discrete Fourier transform (Discrete Fourier Transform), Proposed (DML) represents the first departure angle estimation model designed based on the deterministic maximum likelihood estimation principle in this application, and Proposed (SML) represents the second departure angle estimation model designed based on the stochastic maximum likelihood estimation principle in this application.

[0167] pass Figure 8 It can be seen that the first departure angle estimation model proposed in the present application can accurately estimate the departure angle in a very short sampling time, while other methods or models require more sampling time, that is, collecting more observation sequence data, to achieve similar accuracy. The second departure angle estimation model designed based on the principle of random maximum likelihood estimation proposed in the present application can also obtain higher accuracy after a certain sampling time. Accurate departure angle estimation can be performed with a shorter sampling time, which means that only a small amount of data is needed to make an accurate departure angle estimation. This can significantly reduce resource pressure for IoT devices with scarce antenna resources and accurately estimate the departure angle information of the signal transmitted by the base station terminal.

[0168] The base station transmission power is related to the signal-to-noise ratio of the received signal. The higher the base station transmission power, the higher the signal-to-noise ratio of the signal received by the user equipment, and the higher the signal quality. Fig. 9It can be seen that the first departure angle estimation model proposed in the present application can obtain a good mean absolute error when the base station transmission power is very low. This means that even when the signal quality is poor, the first departure angle estimation model can still obtain a departure angle estimation result with high accuracy based on the received observation sequence data. This means that for IoT devices with scarce antenna resources, even if they receive signals of poor quality, they can still accurately estimate the departure angle through the first departure angle estimation model. The second departure angle estimation model designed based on the principle of random maximum likelihood estimation proposed in the present application can also obtain higher accuracy as the base station transmission power is increased to a certain level.

[0169] In some embodiments, since both the first departure angle estimation model and the second departure angle estimation model are designed by maximum likelihood estimation, although the second departure angle estimation model outputs one more noise variance information than the first departure angle estimation model, the first departure angle estimation model can actually be designed to output the same noise variance information, but it is not used as a key parameter for training. In this case, the second departure angle estimation model and the first departure angle estimation model are designed the same, so they can be designed in the same architecture, simplifying the design of the neural network model.

[0170] The embodiment of the present application obtains the observation sequence data transmitted by the base station and obtains the pilot information based on the communication protocol to obtain the pilot related information. In response to the pilot related information being specific pilot information, the observation sequence data and the pilot information are estimated by the first departure angle estimation model trained in advance to obtain the target departure angle information; in response to the pilot related information being the pilot statistical information of the pilot information, the observation sequence data and the pilot statistical information are estimated by the second departure angle estimation model trained in advance to obtain the target departure angle information. Therefore, in order to improve the accuracy of departure angle estimation, the present application not only uses the received observation sequence data as the input of the neural network model, but also uses the pilot information as the input. According to the obtained pilot related information, if the pilot related information is specific pilot information, the first departure angle estimation model is used to estimate the departure angle. If the pilot related information is pilot statistical information, the second departure angle estimation model is used to estimate the departure angle, so that the user equipment can estimate the departure angle according to different states of the pilot information and obtain accurate target departure angle information.

[0171] See also Fig.10 The embodiment of the present application further provides a departure angle estimation device based on unsupervised learning, which can implement the above-mentioned departure angle estimation method based on unsupervised learning, and the device includes:

[0172] An observation data acquisition module is used to obtain observation sequence data transmitted from the base station;

[0173] A pilot information acquisition module is used to acquire pilot information based on a communication protocol to obtain pilot related information;

[0174] A first angle estimation module is used to estimate the departure angle of the observation sequence data and the pilot information by using a pre-trained first departure angle estimation model to obtain target departure angle information in response to the pilot related information being specific pilot information;

[0175] The second angle estimation module is used to estimate the departure angle of the observed sequence data and the pilot statistical information by using a pre-trained second departure angle estimation model to obtain target departure angle information in response to the pilot related information being the pilot statistical information of the pilot information.

[0176] The specific implementation of the departure angle estimation device based on unsupervised learning is basically the same as the specific implementation of the departure angle estimation method based on unsupervised learning mentioned above, and will not be repeated here.

[0177] The embodiment of the present application also provides an electronic device, the electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the above-mentioned departure angle estimation method based on unsupervised learning when executing the computer program. The electronic device can be any intelligent terminal including a tablet computer, a car computer, etc.

[0178] See also Fig.11 , Fig.11 The hardware structure of an electronic device of another embodiment is illustrated, and the electronic device includes:

[0179] The processor 1101 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application;

[0180] The memory 1102 may be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 1102 may store an operating system and other application programs. When the technical solution provided in the embodiments of this specification is implemented by software or firmware, the relevant program code is stored in the memory 1102, and the processor 1101 calls and executes the departure angle estimation method based on unsupervised learning in the embodiments of this application;

[0181] Input / output interface 1103, used to implement information input and output;

[0182] The communication interface 1104 is used to realize the communication interaction between the device and other devices. The communication can be realized through a wired manner (such as USB, network cable, etc.) or a wireless manner (such as mobile network, WIFI, Bluetooth, etc.);

[0183] A bus 1105 that transmits information between various components of the device (e.g., the processor 1101, the memory 1102, the input / output interface 1103, and the communication interface 1104);

[0184] The processor 1101 , the memory 1102 , the input / output interface 1103 and the communication interface 1104 are connected to each other in communication within the device via the bus 1105 .

[0185] An embodiment of the present application further provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the above-mentioned departure angle estimation method based on unsupervised learning is implemented.

[0186] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0187] The method, device and related medium for estimating the angle of departure based on unsupervised learning provided by the embodiment of the present application obtain the observation sequence data transmitted by the base station and obtain the pilot information based on the communication protocol to obtain the pilot related information. In response to the pilot related information being specific pilot information, the observation sequence data and the pilot information are estimated by the first departure angle estimation model trained in advance to obtain the target departure angle information; in response to the pilot related information being the pilot statistical information of the pilot information, the observation sequence data and the pilot statistical information are estimated by the second departure angle estimation model trained in advance to obtain the target departure angle information. Therefore, in order to improve the accuracy of the departure angle estimation, the present application not only uses the received observation sequence data as the input of the neural network model, but also uses the pilot information as the input. According to the obtained pilot related information, if the pilot related information is specific pilot information, the first departure angle estimation model is used to estimate the departure angle. If the pilot related information is pilot statistical information, the second departure angle estimation model is used to estimate the departure angle, so that the user equipment can estimate the departure angle according to different states of the pilot information and obtain accurate target departure angle information.

[0188] The embodiments described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0189] Those skilled in the art will appreciate that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0190] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0191] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.

[0192] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0193] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0194] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the above units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. The mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0195] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0196] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0197] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including multiple instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM), disk or optical disk and other media that can store programs.

[0198] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but the scope of the rights of the present invention is not limited thereto. Any modification, equivalent substitution and improvement made by a person skilled in the art without departing from the scope and essence of the present invention should be within the scope of the rights of the present invention.

Claims

1. A departure angle estimation method based on unsupervised learning, characterized in that: The method comprises: Obtaining observation sequence data transmitted from a base station; Acquire pilot information based on the communication protocol to obtain pilot related information; In response to the pilot-related information being specific pilot information, performing departure angle estimation on the observation sequence data and the pilot information by using a pre-trained first departure angle estimation model to obtain target departure angle information; In response to the pilot-related information being pilot statistical information of the pilot information, a pre-trained second departure angle estimation model is used to perform departure angle estimation on the observation sequence data and the pilot statistical information to obtain target departure angle information.

2. The method according to claim 1, characterized in that The observation sequence data includes a first number of observation values; The step of estimating the departure angle of the observation sequence data and the pilot information by using a pre-trained first departure angle estimation model to obtain target departure angle information includes: Based on the pilot information, construct a pilot signal matrix; Generate a first model input tensor according to the pilot signal matrix and a first number of the observation values; Inputting the first model input tensor into the first pre-trained departure angle estimation model to perform departure angle estimation to obtain target departure angle information; The step of estimating the angle of departure of the observation sequence data and the pilot statistical information by using a pre-trained second angle of departure estimation model to obtain target angle of departure information includes: Based on the pilot statistical information, construct a pilot vector matrix; Perform Kronecker product processing based on the pilot vector matrix to obtain a pilot processing matrix; generating a second model input tensor according to the pilot processing matrix and the first number of the observations; The second model input tensor is input into the pre-trained second departure angle estimation model to perform departure angle estimation to obtain target departure angle information.

3. The method according to claim 1, characterized in that Before estimating the departure angle of the observation sequence data and the pilot information by using a pre-trained first departure angle estimation model to obtain target departure angle information, the method further includes pre-training the first departure angle estimation model, including: Acquire a first training data set; the first training data set includes first observation training data and pilot training data; Inputting the first observed training data and the pilot training data into an original first departure angle estimation model, and outputting first predicted departure angle information and a first channel gain estimation value; Based on deterministic maximum likelihood estimation, unsupervised learning is performed on the original first angle of departure estimation model according to the first predicted angle of departure information and the first channel gain estimation value to obtain the pre-trained first angle of departure estimation model.

4. The method according to claim 3, characterized in that The method of performing unsupervised learning on the original first angle of departure estimation model based on the deterministic maximum likelihood estimation according to the first predicted angle of departure information and the first channel gain estimation value to obtain the pre-trained first angle of departure estimation model includes: Calculate a steering vector based on the first predicted departure angle information to obtain a first predicted angle steering vector; Constructing a first model loss function based on the pilot training data, the first prediction angle steering vector, the first channel gain estimate and the first observation training data: Wherein, X1 represents the pilot training data, Y1 represents the first observed training data, D1 represents the first training data set, W1 represents the first model parameter of the original first departure angle estimation model, ξ1 represents the first channel gain estimation value, θ1 represents the first predicted departure angle information, and a(θ1) represents the first predicted angle steering vector; The first model parameters of the original first departure angle estimation model are adjusted based on the first model loss function to obtain the pre-trained first departure angle estimation model.

5. The method according to claim 1, characterized in that Before performing departure angle estimation on the observation sequence data and the pilot statistical information by using a pre-trained second departure angle estimation model to obtain target departure angle information, the method further includes pre-training the second departure angle estimation model, including: Acquire a second training data set; the second training data set includes second observed training data and pilot training statistical data; Inputting the second observed training data and the pilot training statistical data into an original second departure angle estimation model, and outputting second predicted departure angle information, a second channel gain estimation value and noise variance information; generating a reconstructed covariance matrix based on the second predicted departure angle information, the second channel gain estimate and the noise variance information, and generating a sample covariance matrix based on the second observed training data; Based on stochastic maximum likelihood estimation, unsupervised learning is performed on the original second departure angle estimation model according to the reconstructed covariance matrix and the sample covariance matrix to obtain a pre-trained second departure angle estimation model.

6. The method according to claim 5, characterized in that The step of generating a reconstructed covariance matrix based on the second predicted departure angle information, the second channel gain estimation value, and the noise variance information, and generating a sample covariance matrix based on the second observed training data includes: Calculate a steering vector based on the second predicted departure angle information to obtain a second predicted angle steering vector; Generate a reconstructed covariance matrix based on the second prediction angle steering vector, the second channel gain estimation value and the noise variance information: Among them, C y represents the reconstructed covariance matrix, ξ2 represents the second channel gain estimate, w (q) represents the noise variance information, θ2 represents the second predicted departure angle information, a(θ2) represents the second predicted angle steering vector, X (q) (X (q) ) H represents the covariance of the pilot information in the pilot training statistical data, It means to find the expected value; Generate a sample covariance matrix based on the second observed training data: in, represents the sample covariance matrix, Q represents the number of the second observed training data, y (q) represents the qth data of the second observed training data, (y (q) ) H Represents y (q) The conjugate transposed matrix of .

7. The method according to claim 5, characterized in that The method of performing unsupervised learning on the original second departure angle estimation model based on the stochastic maximum likelihood estimation according to the reconstructed covariance matrix and the sample covariance matrix to obtain the pre-trained second departure angle estimation model specifically includes: Construct a second model loss function based on the reconstructed covariance matrix and the sample covariance matrix: Among them, C y represents the reconstructed covariance matrix, represents the sample covariance matrix, X2 represents the pilot training statistical data, Y2 represents the second observed training data, D2 represents the second training data set, W2 represents the second model parameter of the original second departure angle estimation model, tr() refers to the trace of the matrix, that is, the sum of the diagonal elements of the main diagonal of the matrix, det(C y ) represents the determinant value of the reconstructed covariance matrix; The second model parameters of the original second departure angle estimation model are adjusted based on the second model loss function to obtain a pre-trained second departure angle estimation model.

8. A departure angle estimation device based on unsupervised learning, characterized in that: The device comprises: An observation data acquisition module is used to obtain observation sequence data transmitted from the base station; A pilot information acquisition module is used to acquire pilot information based on a communication protocol to obtain pilot related information; A first angle estimation module is used for, in response to the pilot-related information being the specific pilot information, estimating the departure angle on the observation sequence data and the pilot information by using a pre-trained first departure angle estimation model to obtain target departure angle information; The second angle estimation module is used to estimate the departure angle of the observation sequence data and the pilot statistical information by using a pre-trained second departure angle estimation model to obtain target departure angle information in response to the pilot related information being the pilot statistical information of the pilot information.

9. An electronic device, characterized in that: The electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the departure angle estimation method based on unsupervised learning as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the departure angle estimation method based on unsupervised learning according to any one of claims 1 to 7 is implemented.

Images