An evaluation method and related device

Abstract

The application provides an evaluation method and related equipment to solve the problem of low estimation accuracy of the to-be-evaluated object in the existing dispatch-free access scene. The method comprises the following steps: obtaining a symbol matrix and an effective angle domain channel matrix; the symbol matrix is constructed based on the symbols sent by the user and received by the antenna array of the base station; the effective angle domain channel matrix is constructed based on the angle domain channel vector from the user to the base station and the active indication parameter set for the user; evaluating the to-be-evaluated object based on a bilinear observation model; the to-be-evaluated object includes at least one of whether the user is active, the signal of the user and the channel from the user to the base station; the bilinear observation model is constructed based on the symbol matrix, the effective angle domain channel matrix and the received signal of the base station for the user.

Description

Technical Field

[0001] This invention relates to the field of wireless communication, and more specifically to an evaluation method, apparatus, computing device, computer-readable storage medium, and computer program product. Background Technology

[0002] With the rapid development of the Internet of Things (IoT) and 5G / 6G communication technologies, large-scale device access has become a significant challenge for wireless communication systems. In real-world applications, although the number of potential access devices is enormous, the number of actually active devices at any given time is relatively small, exhibiting typical sparsity characteristics. To address this characteristic, scheduling-free access technology has gained widespread attention as an efficient access method.

[0003] The principle behind dispatch-free access technology is to allow multiple users to communicate directly using the same spectrum resources without base station scheduling, distinguishing the signals of different users through specific signal processing techniques. Taking its application in the Internet of Things (IoT) as an example, its implementation process includes the following key steps: Active User Identification: The base station needs to identify which IoT devices are active, i.e., devices that are transmitting data. Space-frequency domain-based activity detection algorithms can obtain both coarse and reliable estimates of the active user set. For example, active users can be identified by processing and analyzing received signals and utilizing characteristics such as the sparse structure of the channel matrix.

[0004] Channel State Information (CSI) estimation: Some scheduling-free access technologies require CSI estimation. In some traditional scheduling-free schemes, active devices need to transmit pilot signals for channel estimation so that the receiver can better understand the channel characteristics and thus accurately demodulate the signal.

[0005] Signal detection and separation: The receiver uses advanced signal processing algorithms to detect and separate signals from different IoT devices.

[0006] In existing technologies, numerous Approximate Message Passing (AMP) algorithms based on Bayesian theory have been proposed to solve signal processing problems in large-scale communication scenarios. However, practical experience has shown that traditional AMP methods, when applied to scheduling-free access, suffer from low estimation accuracy for the objects to be evaluated. The objects to be evaluated here include at least one of the following: user activity, user signal strength, and the channel from the user to the base station. Summary of the Invention

[0007] This application provides an evaluation method and related equipment to solve the problem of low estimation accuracy of the object to be evaluated in the prior art in the non-scheduling access scenario.

[0008] The embodiments of this application adopt the following technical solutions: An evaluation method includes: acquiring a symbol matrix and an effective angle-domain channel matrix; the symbol matrix is ​​constructed based on symbols transmitted by a user and received by the antenna array of a base station; the effective angle-domain channel matrix is ​​constructed based on the angle-domain channel vector from the user to the base station and an activity indication parameter set for the user; evaluating an object to be evaluated based on a bilinear observation model; the object to be evaluated includes at least one of whether the user is active, the user's signal, and the channel from the user to the base station; the bilinear observation model is constructed based on the symbol matrix, the effective angle-domain channel matrix, and the received signal from the base station for the user.

[0009] An evaluation apparatus includes: an acquisition unit for acquiring a symbol matrix and an effective angle-domain channel matrix; the symbol matrix is ​​constructed based on symbols transmitted by a user and received by the antenna array of a base station; the effective angle-domain channel matrix is ​​constructed based on the angle-domain channel vector from the user to the base station and an activity indication parameter set for the user; and an evaluation unit for evaluating an object to be evaluated based on a bilinear observation model; the object to be evaluated includes at least one of whether the user is active, the user's signal, and the channel from the user to the base station; the bilinear observation model is constructed based on the symbol matrix, the effective angle-domain channel matrix, and the received signal from the base station for the user.

[0010] A computing device includes: a memory and a processor, wherein the memory is used to store a computer program; and the processor is coupled to the memory and used to execute the computer program stored in the memory to perform the method described above.

[0011] A computer-readable storage medium storing a computer program that, when executed by a computer, enables the implementation of the above-described method.

[0012] A computer program product storing instructions that, when executed by a computer, cause the computer to perform the method described above.

[0013] The at least one solution provided in the embodiments of this application can achieve the following technical effects: Based on a bilinear observation model, the symbol matrix, effective angle domain channel matrix, and base station received signal are deeply coupled. This model can filter redundant interference through the sparsity of the effective angle domain channel matrix and the activity indicator parameter, avoiding error propagation at the modeling level and thus improving the accuracy of the assessment. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the frame structure of the signal transmitted by the user in an embodiment of this application; Figure 2 This is a schematic diagram of the matrix representation of the received signal according to Equation [7] in the embodiments of this application; Figure 3 A schematic diagram illustrating the transmission of signals from a user to a base station; Figure 4 A schematic diagram of the sparse structure of an angle-domain channel; Figure 5 This is a schematic diagram illustrating the specific implementation process of an evaluation method provided in an embodiment of this application; Figure 6 This is a schematic diagram of the specific structure of an evaluation device provided in an embodiment of this application; Figure 7 This is a schematic diagram of the specific structure of a computing device provided in an embodiment of this application. Detailed Implementation

[0015] To address the issue of low estimation accuracy for the object to be evaluated in existing scheduling-free access scenarios, the inventors conducted the following research: The BiGAMP algorithm, or Bilinear Generalized Approximate Message Passing algorithm, is a high-dimensional bilinear model inference algorithm proposed in 2014 by Parker, Schniter, and others. It is an extension of GAMP (Generalized Approximate Message Passing) to bilinear problems. Its core is used for low-rank matrix recovery and factorization, and it has both low complexity and theoretical optimality.

[0016] The essence of the BiGAMP algorithm is an approximate implementation of sum-product belief propagation in the high-dimensional limit. It simplifies message passing through the "Gaussian message approximation" and the "Taylor series approximation" to adapt to the bilinear observation model. Where A and B are the matrices to be recovered, and W is noise. This is an observation mapping. In other words, the BiGAMP algorithm can be used to recover the low-rank matrices A and B from this bilinear observation model.

[0017] Starting from the problem that "the traditional AMP method has low estimation accuracy for the object to be evaluated, which needs further improvement" in existing technologies, and considering the characteristic that "most potential users do not actually access" in large-scale MIMO random access scenarios, the inventors proposed using a "user activity matrix" to perform bilinear modeling together with the channel matrix. Each matrix element in this matrix is ​​an "activity indicator parameter," which characterizes whether the corresponding user is an active user.

[0018] Based on the above analysis, this application proposes to model the "channel matrix" / "user activity matrix" as a model conforming to the "bilinear observation model structure," and then achieve low-complexity, high-precision joint estimation through BiGAMP. Specifically, the detailed demonstration process for modeling the "channel matrix" / "user activity matrix" as a model conforming to the "bilinear observation model structure" is as follows: In this embodiment, it is assumed that in a system with dispatch-free access, each user (equivalent to a user terminal in this embodiment) has Each base station is equipped with a single antenna. A uniform linear array of antennas. For users in this system... The base station receives data from the user. The signal sent It can be represented as: [1] In the above formula: Indicates user The One symbol; Indicates user Access to the base station channel (corresponding to the user) (Channel vector to base station). Indicates user The corresponding additive white Gaussian noise.

[0019] Generally, in a large-scale access scenario, among a large number of potential users, only a small portion are active, and the probability of each user accessing the base station is... They are independent and identically distributed and Based on this characteristic, and combined with the aforementioned method of using a "user activity matrix" to perform bilinear modeling together with the channel matrix, where each matrix element is an "activity indicator parameter" used to characterize whether a corresponding user is an active user—in this embodiment of the application, the activity indicator parameter is used. Indicates user activity status: If the user If it is active, then If inactive, then Therefore, the user activity probability distribution can be expressed as: [2] The meaning of the above formula is: If user If it is active, then the activity indicator parameter The probability of this user being active (i.e., the probability of accessing the base station). = ; and if the user If it is inactive, then the activity indicator parameter The probability of this user being active (i.e., the probability of accessing the base station). = .

[0020] When a user needs the base station to obtain its complete uplink channel matrix, it will activate all N single antennas to transmit pilot signals. Based on this, and according to the above equations [1] and [2], it can be seen that the base station receives the pilot signals from users in an active state. n (correspondingly, ) transmitted signal It can be represented as: [3] In the above formula: This is the signal matrix received by the base station, with dimensions of "number of base station antennas × symbol length"; For users n Active indicator parameters; N The total number of users in the system; For users n The conjugate transpose of the channel vector to the base station is used to describe the user. n Channel response to the base station; This represents the additive white Gaussian noise matrix at the base station receiver, with dimensions equal to... Consistency characterizes noise interference during the reception process; ( ) represents user n Transmitted pilot symbols and transmission symbols—such as Figure 1 The diagram shows the frame structure of the signal transmitted by the user. As can be seen from the diagram, since it is a "scheduling-free access" scenario, the transmission symbols (data) are sent directly to the base station along with the pilot symbols, without the need for communication scheduling by the base station.

[0021] In other words, From pilot symbols and transmission symbols Composition, that is , a superscript p This indicates that the symbol is a pilot symbol. Unique and not orthogonal a superscript d This indicates that the symbol is a transmission symbol. The length of the transmitted symbol is used It means that, among them, for The length of all pilot symbols in the text, for The length of all transmitted symbols in the code.

[0022] Furthermore, the inventors conducted the following research on the "angle domain channel": Angle-domain channel is a channel form in wireless communication that models and represents channel characteristics from an angular dimension. It is a refined perspective for describing spatial domain channels. Its core principle is to use the angle of departure (AoD) of the electromagnetic wave at the transmitting end and the angle of arrival (AoA) of the electromagnetic wave at the receiving end as core parameters to map the continuous channel characteristics in the traditional spatial domain to a discrete or continuous angular dimension, thereby accurately describing the angular distribution of multipath signals.

[0023] In large-scale communication scenarios, the channel can adopt a block fading model, which follows an independent quasi-static flat fading pattern in each coherent time block. The explanation of quasi-static flat fading is as follows: Generally, multiple signals arrive at the receiver at different times, i.e., there is a relative time delay. If these relative delays are much smaller than the time of a symbol, it can be considered that the multiple signals arrive at the receiver almost simultaneously. In this case, multipath fading does not cause inter-symbol interference. This type of fading is called flat fading because the frequency response of such a channel is flat within the used frequency band.

[0024] Conversely, if the relative delays of multiple signals are not negligible compared to the time of a single symbol, then when multiple signals are superimposed, symbols from different times will overlap, causing inter-symbol interference. This type of fading is called frequency-selective fading because the frequency response of such a channel is not flat within the frequency band used.

[0025] Quasi-static flat fading refers to the fact that multipath events do not cause inter-symbol interference, and the fading is constant within each transmission block—this characteristic of constant fading within each transmission block is called quasi-static.

[0026] If the channel adopts a block fading model, then the user Spatial channel vector to base station It can be represented as: [4] In the above formula: Indicates user n The spatial channel vector to the base station; This indicates the bilateral bandwidth, i.e., the frequency range of the channel; Indicates from user The total number of channels to the base station; For users The sequence number of a certain channel to the base station within the total number of channels, with a value range of [1, ..., ... ]; To the user The first to the base station The path complex gain of the channel; for user The first to the base station Path propagation delay of the channel; j Represents the imaginary unit; The normalized angle parameter, i.e., the path incident angle. Normalized representation; The frequency domain phase shift of the multipath component is determined by... and Joint decision; This is the array steering response vector (a vector describing the spatial response characteristics of an antenna array to a signal with a specific incident direction), i.e., the base station antenna array's response to a path incident angle of θ. The spatial response of the signal is calculated using the following formula: [5] In the above formula: , is the normalized angle parameter, i.e., the path incident angle. Normalized representation; For the propagation wavelength, The physical antenna spacing (corresponding to the normalized antenna spacing is) ); The path angle of incidence; M represents the number of array elements contained in the base station antenna array.

[0027] It should be noted that for an antenna array consisting of M elements, when a signal is incident from a certain direction, the signal received by the v-th element in the antenna array will have a phase difference compared to the signal of the 0th element (reference element). The array turning response vector is a vector formed by arranging these phase differences of the M elements in a complex exponential form.

[0028] Next, the active indicator parameters and the corresponding angle-domain channel matrix The product is expressed as: [6] Based on equation [6] and the channel model, the matrix expression for the user's received signal at the base station is transformed into: [7] In the above formula: , represents the symbol matrix received by the base station consisting of symbols (as mentioned above, including pilot symbols and transmission symbols) sent by all users (N users); This represents the activity indicator parameter for all users (N users). and the corresponding angle-domain channel matrix The product of . Considering the characteristics of random user access, Row sparsity. Row sparsity refers to the structural characteristic of a matrix where the vast majority of row vectors are all zero vectors, and only a few row vectors contain non-zero elements.

[0029] In this embodiment of the application, the reason for representing the base station's received signal for the user based on "a matrix expression containing an angle domain signal matrix [7]" is as follows: In a specific example, such as Figure 2 The diagram shown in the matrix representation of the received signal according to Equation [7] is unfolded in matrix form. The horizontal dimension represents the M discrete angle grids of the base station angle domain (determined by the number of base station antennas, corresponding to the column dimension of the effective angle domain channel matrix), and the vertical dimension represents the N potential users in the system (corresponding to the row dimension of the effective angle domain channel matrix). The cells with different filling states in the matrix have clear meanings: the cells filled with color indicate that the effective angle domain channel element corresponding to the "user-angle" position is non-zero, that is, the user has an effective channel response in that angle direction (the signal propagates from that angle to the base station), and the user is an active user (its active indication parameter). =1); an unfilled blank cell indicates that the effective angle field channel element at the corresponding "user-angle" position is zero, which may indicate that the user is inactive. =0), or it could mean that there is no signal propagation in that angular direction for active users (angle domain channel sparsity).

[0030] While the sparsity of the channel matrix helps reduce the computational complexity of channel estimation, a row-sparse structure can enhance the correlation between channels. Specifically, if the elements corresponding to a certain row in the channel matrix... A non-zero value indicates that the user associated with that row is active, and all their channel elements... All of them have the characteristic of taking non-zero values.

[0031] Considering channel correlation, if the base station processes each channel element independently, it will inevitably introduce a large amount of redundant information—that is, for "some channel elements are non-zero, and the rest are zero (e.g., ...)". Not equal to 0, In sparse scenarios where the value is 0, the single-element processing mode will perform invalid calculations on zero elements, resulting in a waste of computing power.

[0032] Conversely, if the base station directly uses the complete channel row vector While processing data as a whole can avoid interference from redundant information, the computational complexity of signal processing increases exponentially because the row vector dimension is the same as the number of potential users, making it difficult to meet the real-time requirements of large-scale MIMO systems.

[0033] To address the aforementioned technical contradictions, the inventors considered adopting a channel preprocessing scheme based on angle domain transformation. That is, by projecting the time-domain / spatial-domain channel matrix to the angle domain, the correlation between channels can be effectively reduced, thereby ensuring estimation accuracy while also achieving low algorithm complexity.

[0034] In practical applications, base stations are typically located in elevated positions with few surrounding scattering objects, while users are usually located at lower altitudes in environments with abundant local scattering, and are generally far from the base station. A schematic diagram of a user transmitting signals to a base station is shown below. Figure 3 As shown: The user sends a signal, via... A path leads to the base station, which receives user signals through antenna beam scanning.

[0035] The spatial domain-based channel model represents the signal received by each antenna in a certain time slot. Therefore, for active users, the spatial domain signal is usually not zero.

[0036] Angle domain transformation, on the other hand, uses Fourier transform to... Space divided M share, MGiven the number of antennas, the signal received at each angle is calculated. Since the propagation of wireless signals is limited by the effective scattering path and direction of arrival (DOA) in physical space, only a few discrete angles correspond to effective propagation paths, while the vast majority of angles receive no signal energy. Therefore, the spatial channel matrix, after angle-domain projection transformation, exhibits significant sparsity. Furthermore, the number of base station receiving antennas... M The larger the value, the higher the resolution of the angle domain, the finer the division of the angle grid, the lower the proportion of the number of non-zero angle grids corresponding to the effective propagation path to the total number of grids, and the higher the sparsity of the angle domain channel matrix.

[0037] Since the transformation relationship between the "spatial domain channel" and the "angle domain channel" is that the spatial domain channel is obtained by the conjugate transpose of the angle domain channel through the transformation matrix, in order to achieve the goal of "projecting the time / spatial domain channel matrix into the angle domain to effectively reduce the correlation between channels", the transformation matrix can be transformed into the angular domain channel matrix. The spatial domain channel corresponding to each user is represented as follows: [8] In the above formula: For users n The corresponding angle domain channel vector has a dimension of M×1, where M is the number of base station antennas; For users n The corresponding spatial domain channel vector, with a dimension of M×1, describes "the channel gain between each antenna and the user"; This is the conjugate transpose of the transformation matrix (the superscript H indicates the conjugate transpose, which is a common operation for complex matrices). It is a transformation matrix, specifically a Discrete Fourier Transform (DFT) matrix, used to perform the transformation from the "spatial domain" to the "angular domain". The positive transformation of corresponds to 'spatial domain → angular domain', and its conjugate transpose corresponds to 'angular domain → spatial domain'. ,in (The value of m ranges from [1, M]) is the base station reception angle, defined as: [9] in, . This is the middle index (when M is odd, it corresponds to the "right middle" angle sampling point).

[0038] m is the index of the angle sampling point, ranging from 1 to M. M represents the number of array elements contained in the base station antenna array.

[0039] Represents the m-th angle The array turning response vector corresponding to (base station reception angle). Wherein, It is a "discrete sampling point" in the angle domain (discretizing a continuous angle space into M points).

[0040] Based on Discrete Fourier Transform Matrix The "received signal of each antenna (spatial domain)" can be converted into the "received signal of each angle (angle domain)". Since the signal only propagates from a few directions, the angle domain channel is sparse (which is also the basis for the subsequent "sparse channel estimation").

[0041] Next, to simplify signal processing by combining "user activity state" and "angle domain sparsity", the user can be... n The product of the activity indication parameter and the corresponding angle-domain channel is expressed as:

[10] In this way, the effective angle-domain channel vector of active users can be obtained. .

[0042] In the above formula: For users n The effective angular domain channel vector (dimension M×1); For users n Active indicator parameters; For users n Angular domain channel vector (previously defined) , here simplified as ).

[0043] Accordingly, based on the channel model, the base station's expression for the user's "received signal in the angle domain" transforms into:

[11] In the above formula: The spatial domain received signal matrix has a dimension of L×M, where L is the number of time slots; , represents the symbol matrix consisting of symbols (including pilot symbols and transmission symbols) transmitted by each antenna of multiple users; N is the total number of users; For example, the symbol sent by the Nth user, and so on; The discrete Fourier transform matrix has dimensions M×M; The effective angular domain channel matrix for users has dimensions N×M; , is composed of the effective angle domain channel vectors of all users as shown in Equation

[10] ; Let represent the spatial domain additive Gaussian noise matrix with dimensions L×M.

[0044] It should be noted that: Active users ( =1) in The corresponding row vector is Preserving the sparse non-zero characteristics of the angle domain channel, it becomes the core data for subsequent identification of active users and estimation of channel gain based on the observation mean of channel elements; Inactive users ( =0) in The corresponding row vectors are all-zero vectors. They do not contribute effective channel information in matrix operations, but only serve as "placeholder rows" to reflect the potential user scale of the system, thus avoiding frequent adjustments to the matrix dimensions due to changes in user activity status.

[0045] like Figure 4 The figure shown is a schematic diagram of the sparse structure of the angle domain channel. Figure 4 In the diagram, the vertical axis represents all users in the system (N in number); the horizontal axis corresponds to M discrete angles in the angle domain (i.e., ... (This is determined by the number of base station antennas). Figure 4 The blue blocks represent "the user's channel is non-zero at the corresponding angle" (i.e., the signal propagates to the base station from this angle); the blank blocks represent "the user's channel is 0 at the corresponding angle" (i.e., no signal propagates). Figure 4 In the diagram, since each user's row contains only a few blue blocks (most are blank), it indicates that a single user's signal propagates from only a few angles, and the corresponding angle domain channel is sparse. After combining the "active indicator parameter", the rows of inactive users are all blank, while the rows of active users contain only a few blue blocks, indicating that the final effective angle domain channel matrix X is a sparse matrix.

[0046] Furthermore, if the received signal in the spatial domain is converted to the received signal in the angular domain, the received signal in the angular domain has the following expression:

[12] The meaning of equation

[12] is: through inverse transformation (transforming...) Multiply This converts the spatial domain received signal into the angular domain received signal.

[0047] In the above formula: The angle-domain received signal matrix has dimensions L×M; , indicating the orthogonality of the discrete Fourier transform matrix; It is an identity matrix, therefore we have ; , represents the symbol matrix consisting of symbols (including pilot symbols and transmission symbols) transmitted by each antenna of multiple users; N is the total number of users; The effective angular domain channel matrix for users has dimensions N×M; It consists of the effective angle domain channel vectors of all users as shown in Equation

[10] . Each row corresponds to a user's effective angle domain channel vector; Let be the additive Gaussian noise matrix in the angle domain. Considering the properties of the Gaussian distribution, It is still additive white Gaussian noise and .

[0048] As can be seen from the above derivation process, the inventors, through research and reasoning, completed the derivation from "active user definition" to "actual received signal in the spatial domain". By transforming the channel to the angular domain, the correlation between channels is reduced, changing from row sparsity to random sparsity. Therefore, sparse recovery methods can be used to jointly identify active users and detect signals.

[0049] Based on the above research and analysis results, in order to solve the problem of low estimation accuracy of the object to be evaluated in the existing scheduling-free access scenario, this application provides a user identification and signal detection method.

[0050] The subject executing this method can be any computing device capable of implementing the method, such as base stations, servers, mobile phones, personal computers, smart wearable devices, smart robots, and so on.

[0051] Different steps of this method can be implemented by the same execution entity or by different execution entities. This application does not limit which execution entity is used to implement the method.

[0052] Furthermore, the embodiments of this application do not limit the execution order of different steps. When using the method provided in the embodiments of this application, the execution order of different steps can be adjusted according to actual needs.

[0053] For ease of description, the following uses a base station as the execution subject of this method as an example to provide a detailed description of the method provided in the embodiments of this application.

[0054] The flowchart of the specific implementation of this method is as follows: Figure 5 As shown, it includes the following steps: Step 11: Obtain the symbol matrix and the effective angle domain channel matrix; The symbol matrix is ​​constructed based on the symbols transmitted by the user and received by the base station's antenna array. The symbols transmitted by the user can consist of pilot symbols and transmission symbols.

[0055] In one alternative implementation, A mentioned above is the symbol matrix described herein.

[0056] In one alternative implementation, the base station can receive the radio symbol stream transmitted by the user through a multi-antenna array, perform down-conversion, analog-to-digital conversion (ADC), and synchronization processing on the received signal to obtain a time-domain received signal sequence. Assuming the base station antenna array has M antennas and the number of receiving time slots is L, the symbols received by each antenna in each time slot are arranged along the "time slot-antenna" dimension to construct a symbol matrix.

[0057] The effective angle domain channel matrix is ​​constructed based on the angle domain channel vector from the user to the base station and the active indication parameters set for the user.

[0058] In an optional implementation, X mentioned above is the effective angle domain channel matrix described in step 11.

[0059] In one alternative implementation, the effective angle-domain channel matrix can be constructed in the following manner: First, obtain the spatial domain channel matrix; the spatial domain channel matrix is ​​constructed based on the spatial domain channel vector corresponding to the user. Then, using the discrete Fourier transform matrix, the spatial domain channel matrix in the spatial domain channel matrix is ​​transformed into the angle domain channel matrix; the angle domain channel matrix contains the angle domain channel vector. Finally, an effective angle-domain channel matrix is ​​constructed based on the angle-domain channel matrix and the activity indicator parameter.

[0060] In one alternative implementation, for each user n, based on the angle response characteristics of the base station antenna array, an angle domain transformation (such as Discrete Fourier Transform DFT or Sparse Bayesian Learning Transform) is performed on the corresponding spatial channel to obtain the angle domain channel vector of user n. The elements in the vector represent the channel gain of user n in the m-th angle direction, which naturally has the characteristic of sparsity (active users correspond to a small number of non-zero elements, while inactive users have elements close to zero).

[0061] An activity indicator parameter is set for each user n, where an activity indicator parameter of 1 indicates that user n is a potentially active user, and 0 indicates that user n is a potentially inactive user. Initially, the activity indicator parameter can be initialized based on historical communication data from the base station or a preset threshold: if the average historical channel gain of the user is greater than the set threshold, the activity indicator parameter is initialized to 1; otherwise, it is initialized to 0. This parameter can be iteratively updated through the active user determination step.

[0062] Arrange all users' angle domain channel vectors according to the "user-angle" dimension to obtain the initial angle domain channel matrix, whose row vectors are the angle domain channel vectors of each user. Then, the effective angle domain channel vectors of each user can be calculated according to Equation

[10] .

[0063] Step 12: Evaluate the object to be evaluated based on the bilinear observation model; The objects to be evaluated may include at least one of the following: whether the user is active, the user's signal strength, and the channel from the user to the base station.

[0064] The bilinear observation model mentioned here is constructed based on the symbol matrix, the effective angle domain channel matrix, and the base station's received signal for the user.

[0065] In an alternative implementation, if the object to be evaluated includes whether a user is active, then step 12 may include: determining whether a user is an active user based on the influence of the symbol matrix and the received signal on the effective angle domain channel matrix in the bilinear observation model.

[0066] In a specific example, the impact results mentioned here may include the observation mean of the channel elements in the effective angle domain channel matrix in the bilinear observation model.

[0067] In an optional implementation, if the influence result is the observed mean, then step 12 may specifically include: In one optional implementation, if the object to be evaluated includes whether a user is active, the posterior probability weight of each channel element is calculated based on the observed mean. Then, the user is determined to be an active user based on the value of the posterior probability weight and a pre-set active user discrimination criterion based on the value of the posterior probability weight.

[0068] In one alternative implementation, for the constructed effective angle domain channel matrix, the mean of all channel element observations for a single user in each row of the matrix can be extracted. The mean of all channel element observations in that row collectively characterizes the channel response features of that user in different angular directions.

[0069] For each user's set of channel element observation means, analyze the numerical characteristics of each element: if at least one of the channel element observation means for the user is significantly different from the noise floor level, then the user's angle domain channel has an effective response; otherwise, if all the channel element observation means for the user are not significantly different from the noise floor level, then the user's angle domain channel has no effective response.

[0070] Users who have a valid response in the angle domain channel are identified as active users, while users who have no valid response in the angle domain channel are identified as inactive users, thus completing the distinction of the active status of all users.

[0071] In an alternative implementation, if the object to be evaluated includes a channel, then step 12 may further include: based on channel elements Posterior probability weights Gaussian distribution variance when channel elements are non-zero Noise variance of observed mean and observed mean The expected value of the marginal posterior probability of the channel element is calculated as an approximate estimation result of the channel element. This part can be found in the following formula

[20] .

[0072] In an optional implementation, if the object to be evaluated includes a user's signal, then step 12 may further include: calculating the expected value of the marginal posterior probability of the signal elements in the symbol matrix in the bilinear observation model, as an approximate estimate of the user's signal. Specifically, refer to Equation

[16] and its related textual description below.

[0073] Specifically, based on Bayesian estimation logic, combined with the observation data of the symbol matrix and the channel response characteristics reflected by the effective angle domain channel matrix, the marginal posterior probability distribution of the target signal element can be constructed. This distribution can integrate the influence of observation noise and the channel sparsity characteristics to accurately characterize the true value probability features of the signal element.

[0074] By performing expectation operation on the constructed marginal posterior probability distribution, the expected value of the distribution is obtained. This expected value is the optimal approximate estimation result of the signal element under the minimum mean square error criterion. It is used as the signal estimation value of the corresponding active user to complete the approximate estimation of the single user signal. The above operation is performed sequentially on the signal elements of the symbol matrix corresponding to all active users to finally obtain the approximate estimation results of the signals of all active users.

[0075] The following section further details the implementation of step 12 and the related model derivation process: Based on the preceding research and analysis, in this embodiment, the "channel matrix" / "user activity matrix" is modeled as a bilinear structure. Thus, we expect to obtain the channel matrix... and user signal matrix The minimum mean square error (MMSE).

[0076] 1) First, to calculate the MMSE of X and A, we can first calculate... and The marginal probability function. Specifically, the BiGAMP algorithm can be used to calculate the marginal posterior probabilities of X and A.

[0077] In the fields of probability statistics and communication signal processing, several concepts related to probability distributions are introduced below: The joint probability distribution of multiple variables is a function that describes the probability of two or more random variables (such as X and A) taking the same value at the same time. It fully describes the interdependence between multiple random variables and is a basic tool for analyzing multivariate systems.

[0078] Marginal probability functions are probability distribution functions calculated for one or more remaining variables in a multivariate joint probability distribution after ignoring some variables (such as ignoring X or A).

[0079] The probability distribution function, also known as the cumulative distribution function, is a function that describes the probability that a random variable will take a value less than or equal to a certain specific value. It is one of the core tools for characterizing the probabilistic laws of random variables and is applicable to all types of random variables, including discrete, continuous, and mixed types.

[0080] Posterior probability refers to the probability inference of the value of an unknown parameter or random variable after observing certain data / events. It is a core concept of Bayesian statistics. Its essence is to combine prior knowledge with observed data to update the probabilistic understanding of unknown variables.

[0081] Marginal posterior probability is the posterior probability of a variable when there are multiple unknown variables (e.g., channel X and signal A are both unknowns). It refers to the probability of a variable obtained by integrating / summing over all other unknown variables when the observed result (e.g., Y) is known, focusing on only one unknown variable.

[0082] The joint posterior probability distribution is an extension of posterior probability in multivariate scenarios. It refers to the probability distribution of multiple unknown random variables taking values ​​simultaneously after data has been observed. It fully characterizes the dependencies between multiple variables and the correlation with observed data, and is a core tool for multivariate joint inference within the Bayesian framework.

[0083] Prior probability distribution is a fundamental concept in Bayesian statistics, referring to the probabilistic understanding of the values ​​of an unknown random variable or parameter before any observed data or experimental results are available. This understanding is usually based on domain experience, physical laws, or mathematical model assumptions, and serves as "prior knowledge" for subsequently deriving posterior probabilities from observed data.

[0084] In this embodiment of the application, according to Bayes' theorem—when the received signal Y is known, the "joint posterior probability of channel X and user signal A" is proportional to "'likelihood probability × prior probability'" ( (This means 'proportional to'), so we know... and The joint posterior probability distribution is:

[13] In the above formula: Let X be the joint posterior probability distribution of X and A, also known as the multivariate joint probability distribution—that is, the probability that X and A simultaneously take a certain value given the received signal Y. Y represents the received signal from the base station; It represents the likelihood probability—that is, the probability of receiving Y given X and A. The prior probability distribution of channel X—that is, the probability distribution of X before the received signal (determined by the channel characteristics)—is as follows: .in, Channel matrix The elements corresponding to "user n, angle m" in the table; This refers to taking the product of the elements of all users n and all angles m (reflecting independence).

[0085] The prior probability distribution of user signal A—that is, the probability distribution of A before the received signal (determined by pilot / signal design)—is as follows: .in, Signal matrix "time slot" l The element corresponding to "user n"; For all time slots l Take the product of all elements of user n (to reflect independence).

[0086] Since the noise W follows a Gaussian distribution, the received signal is expressed according to equation

[12] . and will and Substituting the above specific calculation method into equation

[13] , equation

[13] can be transformed into:

[0087] The above formula means that under Gaussian noise, the likelihood probability follows an exponential distribution (negatively correlated with the squared error between the received signal and the estimated signal).

[0088] In the above formula: This represents the element corresponding to "time slot l, antenna m" in the received signal matrix Y; This is the estimated value of "user signal A × channel X" at "time slot l, antenna m"; The variance of the noise; It is an exponential function, representing the probability density form of a Gaussian distribution.

[0089] In this embodiment, the "channel estimation + active user identification" problem is transformed into a "posterior probability maximization problem" using a Bayesian framework. Based on the above equation, sum-product algorithms and factor graphs can typically be used to obtain... and Marginal posterior probability distribution and However, such an algorithm is too complex. Therefore, in the embodiments of this application, the AMP (Approximate Message Passing) algorithm will be used to simplify it and achieve efficient solution of high-dimensional problems.

[0090] Specifically, in the embodiments of this application, we can use the received signal in the angle domain represented by

[12] Represented as:

[0091] Thus, the received signal model (Equation

[12] ) is decomposed into a linear expression of individual elements. The meaning of this expression is: the received signal at "time slot l, antenna m". , is "all signals of user n" ×channel The superposition of "" plus the noise at that location .

[0092] In the above formula: For the "time slot l, antenna m" elements of the received signal matrix Y; These are the "time slot l, user n" elements of the signal matrix A; These are the "user n, angle m" elements of the channel matrix X; These are the "time slot l, antenna m" elements of the noise matrix w.

[0093] As mentioned above, and The unknown, the received signal is a bilinear combination of "signal × channel", therefore, for and The estimation problem is a bilinear reasoning problem that can be solved using BiGAMP.

[0094] The core idea of ​​the BiGAMP algorithm is to approximate the influence of other variables on the current variable as a noisy Gaussian observation (i.e., a Gaussian message) during the iteration process. For channel elements... Other variables (such as) , The effect of )) will be simplified to " A Gaussian observation model.

[0095] So, for channel elements According to the BiGAMP algorithm, its marginal posterior probability distribution is as follows

[14] :

[14] Equation

[14] is a combination of "Gaussian message approximation" and "Bayes' theorem": BiGAMP simplifies the complex joint posterior probability through iteration, transforming it into the form of "prior distribution × Gaussian observation", thereby efficiently obtaining a single channel element. The marginal posterior probability distribution.

[0096] The meaning of equation

[14] is: the received signal is known At that time, channel element The approximate marginal posterior probability distribution (output of the BiGAMP algorithm).

[0097] In equation

[14] : for Marginal posterior probability distribution (focusing only on) (ignoring other variables); for The prior probability distribution (prior characteristics of the channel). The complex Gaussian distribution is obtained during the iteration process of the BiGAMP algorithm. "Approximate observation model", where: For the t-th iteration, The estimated mean (considered as) (the "observation value"); The noise variance of the observation at the t-th iteration; variable It can be considered real The observed values ​​are also subject to variance of Interference from zero-mean Gaussian noise; The denominator in the above formula is the normalized integral (which guarantees that the sum of the posterior probabilities is 1).

[0098] And for signal elements According to the BiGAMP algorithm, its marginal posterior probability distribution is as follows

[15] :

[15] In the above formula: For signal elements The marginal posterior probability distribution; variable It can be understood as the true value The measured values, where the observed values ​​are varianced by . Zero-mean Gaussian noise interference; yes The prior probability distribution; It is obtained by BiGAMP iteration. "Approximate observation model", where: For the t-th iteration, The estimated mean; This represents the noise variance of the observed value. The denominator in the above formula is the normalized integral (which guarantees that the sum of the posterior probabilities is 1).

[0099] When we obtain the channel After the marginal posterior probability distribution (such as Equation

[14] ), this probability distribution describes " "What values ​​are possible, and what is the probability of each value?" What we ultimately need is... Specific estimated values ​​(e.g., " When asked to determine the approximate value of a probability distribution, the most common approach is to calculate the expected value (mean) of that posterior probability distribution. This is because the expected value is the "average representative value" of the probability distribution, reflecting its approximate value to the greatest extent possible. Possible values ​​for .

[0100] Similarly, for signal elements It is also possible to calculate the expectation of its posterior probability distribution.

[0101] Therefore, the following calculation is performed: and The operation of "expectation of the marginal posterior probability distribution".

[0102] 2) Calculate " and "The expectation of the marginal posterior probability distribution".

[0103] According to the aforementioned and Marginal posterior probability distribution, combined with and The prior distribution can be calculated separately. and The expectation of the posterior probability distribution is used to obtain the MMSE estimate. and MMSE estimates .in, For the (t+1)th iteration, the signal element The MMSE estimate (minimum mean square error estimate, which is one of the optimal estimates); For the (t+1)th iteration, the channel element The estimated value of MMSE.

[0104] In this embodiment of the application, since the pilot sequence base station is known, it is possible to obtain the current... ( When (is the number of pilot time slots), = Considering that the transmitted signal uses QPSK modulation (rather than known pilots), according to equation

[16] , when At that time, the MMSE estimation method for QPSK signals is adopted—utilizing the finiteness of modulation constellation points, the signal estimate is obtained through soft-decision weighting. It can be represented as:

[16] As can be seen above, the applicable scenarios of formula

[16] include: when At that time, the transmitted signal uses QPSK modulation (rather than known pilots), and it is necessary to estimate the signal. .

[0105] In the above formula

[16] : For complex constellation points in QPSK; For the t-th iteration, the signal The observed mean (from the Gaussian approximation of BiGAMP, corresponding to the previous) ); For the t-th iteration, the observed value The noise variance (from the Gaussian approximation of BiGAMP). Soft decision weights are used to measure "constellation points". Compared with observed values The degree of matching (the smaller the distance, the greater the weight).

[0106] The core logic of Equation

[16] is that the QPSK signal can only take 4 constellation points. Therefore, the MMSE estimate of the signal is obtained by “calculating the ‘matching weight’ for each constellation point and then weighting the average” (which is essentially the expectation of the posterior probability).

[0107] The following explains how to obtain channel elements. MMSE estimates : Although The prior distribution exhibits inherent complexity under the channel model defined in Equation [4], but Monte Carlo simulations show that... The distribution of can be precisely approximated by a Gaussian distribution. Without loss of generality, we can assume that . It follows a Bernoulli-Gaussian distribution, and its prior probability distribution is... It can be represented as follows:

[17] The above formula

[17] is applicable to the following scenarios: angle domain channels. It is sparse, therefore it needs to be described using a sparse prior distribution. According to

[17] , that is, using the Bernoulli-Gaussian distribution to describe It has the sparse property of "most elements are 0 and a few are Gaussian values".

[0108] In the above formula

[17] : For channel elements The prior probability distribution; for The sparse probability (i.e., non-zero probability) represents The probability is 0 (usually very small because the channel is sparse); due to the angle domain transformation of the channel, Usually less than ; Represents the impulse function—when When =0, =1; for The probability distribution when the value is non-zero—as assumed above—is a complex Gaussian distribution. .

[0109] The core logic of equation

[17] is: With probability Taking 0 (sparseness), with probability Take the non-zero values ​​of the complex Gaussian distribution (the actual characteristics of the channel), which is the "Bernoulli-Gaussian distribution" (the available prior for sparse signals).

[0110] According to Bayes' theorem, the posterior probability formula The distribution of the marginal posterior probability can be expressed as the ratio of "likelihood × prior", specifically as follows:

[18] Equation

[18] is "channel element" Non-zero posterior probability weights (measuring channel elements) The probability is not zero.

[0111] In formula

[18] : For "channel element" The non-zero posterior probability weights are auxiliary functions that combine observations and priors, and are channel elements. A quantitative indicator of "non-zero probability"; for The prior non-zero probability (i.e., sparse probability). The likelihood ratio is obtained from observational information; among which... As shown below

[19] :

[19] In formula

[19] : For observations The noise variance; for The variance of the Gaussian distribution when it is non-zero (prior); for The observed mean, that is, the mean of the observations. Noisy measurements; for The conjugate transpose (the inner product operation of complex numbers).

[0112] It should be noted that in bilinear inference, the channel... and signal All of these are unknown; the BiGAMP algorithm uses "message passing" to pass on "other variables (such as...)". Received signal )right The "impact" can be simplified into a noisy "virtual observation"—the "central value" of this virtual observation is... That is, the "observation mean".

[0113] The channel element shown in equation

[17] Prior probability distribution , and the weighting factor of the sparse probability shown in Equation

[18] Substituting into equation

[14] , we can calculate... The expected value of the marginal posterior probability (i.e., the MMSE estimate) ), as shown in the following formula

[20] :

[20] Equation

[20] is the channel element. The method for calculating the MMSE estimate (expectation of posterior probability).

[0114] In formula

[20] : This represents the expectation operation (the expectation of the marginal posterior probability, i.e., the MMSE estimate—the estimate that minimizes the MMSE is exactly the expectation of the unknown variable under the marginal posterior probability distribution). for The variance of the Gaussian distribution when it is non-zero (prior); “ “ in ", indicating if Determine the posterior mean of a non-zero Gaussian distribution.

[0115] Considering the calculation according to formula

[18] When, in formula

[18] sparse probability Typically unknown, in this embodiment, we use the EM algorithm to iteratively approximate the parameters. The EM algorithm is an iterative algorithm used for maximum likelihood estimation or maximum a posteriori probability estimation of parameters in probabilistic models containing latent variables. Each iteration of the EM algorithm consists of two steps: the E-step, which calculates the expectation; and the M-step, which calculates the maximization.

[0116] The following describes how the EM algorithm is used to approximate the embodiments of this application. For the sake of brevity, the number of iterations will be omitted in this part of the derivation. Specifically, we will first assume an initial... (i.e., the sparse probability of the j-th iteration); then, calculate using the aforementioned formula. and Then, update .

[0117] As mentioned above, we first assume that the sparsity of the j-th EM update has already been obtained. Therefore, the EM recursion can be written as: [twenty one] The meaning of formula

[21] is: The EM algorithm consists of two steps (E-step + M-step): E-step: Fix the current sparsity probability (Result of the j-th iteration), calculate the "log prior probability" Regarding the posterior distribution The expected value (i.e., the integral term); M-step: Find the step that maximizes this expectation. As the next iteration .

[0118] Substituting the channel prior (Bernoulli-Gaussian distribution) and posterior distribution from the previous text into

[21] , and by calculating... The outpost (the outpost is unique),

[21] can be simplified and updated as follows: [twenty two] As shown in

[22] , the sparse probability of the next iteration This is equivalent to the "channel non-zero posterior probability weight" mentioned earlier. (using the current) (Calculated).

[0119] The following describes the criteria for identifying active users and their signal detection: Assumption express Not 0, otherwise It can be proven that: [twenty three] That is, "known observations" hour, active( The posterior probability of ) — therefore Can be directly characterized Activity level, specifically: when hour: High probability of non-zero (channel has value) → corresponds to "user n is active at angle m"; when hour: The probability is zero (no channel value) → corresponding to "user n is inactive at angle m".

[0120] Through the above process, we can obtain: Channel estimation (According to formula

[20] ); Signal estimation ; User activity estimation .

[0121] If we define the criteria for identifying active users as follows:

[0122] Therefore, according to this criterion, when At that time, we believe users It is active; otherwise, the user is considered active. Inactive.

[0123] In other words, in one alternative implementation, the mean of the observed channel elements in the effective angle domain channel matrix can be used as the basis. Calculate the posterior probability weights of each channel element for non-zero channel elements. The value of the posterior probability weight is used to determine whether the user corresponding to the channel element in the effective angle domain channel matrix is ​​an active user, based on the value of the posterior probability weight and the pre-set active user discrimination criterion based on the value of the posterior probability weight.

[0124] Assuming user If it is active, then its signal is:

[0125] in, It is calculated according to the aforementioned formula

[16] : "time slot" l ,user k The signal is the MMSE estimation result after the t-th iteration. Since user k is a real active user, and It is a signal estimate calculated for user k, therefore the obtained value is directly used. As a signal of active user k, thus having .

[0126] As can be seen, in the embodiments of this application, it is possible to first use... Determine if user k is active; if active, directly use the signal estimate obtained from the previous iteration. This serves as a signal for the user.

[0127] The method described in this application, based on a bilinear observation model, deeply couples the symbol matrix, the effective angle domain channel matrix, and the base station received signal. This model can filter redundant interference through the sparsity characteristics of the effective angle domain channel matrix and the activity indicator parameter, thereby avoiding error propagation at the modeling level and improving the accuracy of the evaluation.

[0128] Furthermore, active user identification is based on the observed mean of the effective angle domain channel matrix. This mean is directly related to the user's angle domain channel characteristics and can clearly distinguish the non-zero channel response of active users from the noise components of inactive users, avoiding misjudgment and missed judgment by simply relying on signal strength. Signal detection uses the expected value of the edge posterior probability. Compared with the hard decision and linear estimation of existing technologies, this method has theoretical optimality under the mean square error criterion and can approximate the true signal value to the greatest extent, thereby improving detection accuracy and precision.

[0129] Example 2 To address the problem of low estimation accuracy of the object to be evaluated in the prior art in the scheduling-free access scenario, and based on the same inventive concept as the above embodiments of this application, Embodiment 2 of this application provides an evaluation device.

[0130] A schematic diagram of the specific structure of the device is shown below. Figure 6 As shown, it includes the following functional units: Acquisition unit 61 is used to acquire the symbol matrix and the effective angle domain channel matrix; The symbol matrix is ​​constructed based on the symbols sent by the user and received by the base station's antenna array; the effective angle domain channel matrix is ​​constructed based on the angle domain channel vector from the user to the base station and the active indication parameter set for the user.

[0131] Evaluation unit 62 is used to evaluate the object to be evaluated based on a bilinear observation model.

[0132] The objects to be evaluated include at least one of the following: whether the user is active, the user's signal strength, and the channel from the user to the base station.

[0133] The bilinear observation model is constructed based on the symbol matrix, the effective angle domain channel matrix, and the received signal from the base station for the user.

[0134] In one optional implementation, the object to be evaluated includes the user's signal; then, the evaluation unit 62 can be specifically used to: calculate the expected value of the marginal posterior probability of the signal element in the symbol matrix in the bilinear observation model, as an approximate estimation result of the user's signal.

[0135] In an optional implementation, if the object to be evaluated includes whether the user is active, then the evaluation unit 62 can be specifically used to: determine whether the user is an active user based on the influence of the symbol matrix and the received signal on the effective angle domain channel matrix in the bilinear observation model.

[0136] In an optional implementation, if the influence result includes the observation mean of the channel elements in the effective angle domain channel matrix in the bilinear observation model, then the evaluation unit 62 can be specifically used to: calculate the value of the posterior probability weight of each channel element when the channel element is non-zero based on the observation mean; and determine whether the user is an active user based on the value of the posterior probability weight and a pre-set active user discrimination criterion based on the value of the posterior probability weight.

[0137] In an optional implementation, if the influence result includes: the observation mean of the channel element in the effective angular domain channel matrix in the bilinear observation model; and the object to be evaluated includes the channel; then, the evaluation unit 62 can specifically be used to: calculate the expected value of the marginal posterior probability of the channel element based on the posterior probability weight of the channel element, the Gaussian distribution variance when the channel element is non-zero, the noise variance of the observation mean, and the observation mean, as an approximate estimation result of the channel element.

[0138] In one optional implementation, the apparatus provided in this application embodiment may further include: A spatial domain channel matrix acquisition unit is used to acquire a spatial domain channel matrix; the spatial domain channel matrix is ​​constructed based on the spatial domain channel vector corresponding to the user. The transformation unit is used to convert the spatial domain channel matrix into an angle domain channel matrix using a discrete Fourier transform matrix; the angle domain channel matrix contains the angle domain channel vector. The construction unit is used to construct the effective angle domain channel matrix based on the angle domain channel matrix and the active indication parameter.

[0139] The apparatus described in this application deeply couples the symbol matrix, the effective angle domain channel matrix, and the base station received signal by relying on a bilinear observation model. This model can filter redundant interference through the sparsity characteristics of the effective angle domain channel matrix and the activity indicator parameter, thereby avoiding error propagation at the modeling level and improving the accuracy of the evaluation.

[0140] Furthermore, active user identification is based on the observed mean of the effective angle domain channel matrix. This mean is directly related to the user's angle domain channel characteristics and can clearly distinguish the non-zero channel response of active users from the noise components of inactive users, avoiding misjudgment and missed judgment by simply relying on signal strength. Signal detection uses the expected value of the edge posterior probability. Compared with the hard decision and linear estimation of existing technologies, this method has theoretical optimality under the mean square error criterion and can approximate the true signal value to the greatest extent, thereby improving detection accuracy and precision.

[0141] Example 3 Based on the same inventive concept as the foregoing embodiments of this application, Embodiment 3 of this application provides a computing device to solve the problem of low estimation accuracy of the object to be evaluated in the prior art in the non-scheduling access scenario.

[0142] like Figure 7 As shown, the computing device includes a memory 71 and a processor 72. The memory 71 can be configured to store various other data to support operation on the electronic device. Examples of this data include instructions for any application or method used to operate on the electronic device. The memory 71 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0143] The processor 72, coupled to the memory 71, is used to execute a computer program stored in the memory 71 for performing an evaluation method as described in Embodiment 1 of this application.

[0144] When the processor 72 executes the computer program in the memory 71, in addition to the functions described above, it can also perform other functions, as detailed in the descriptions of the preceding embodiments.

[0145] Furthermore, such as Figure 7 As shown, the computing device also includes other components such as a display 74, a communication component 73, a power supply component 75, and an audio component 76. Figure 7 The diagram only shows some components and does not mean that the computing device includes only these components. Figure 7 The components shown.

[0146] Accordingly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed by a computer, can implement the methods provided in the above embodiments.

[0147] Accordingly, this application also provides a computer program product, which stores instructions that, when executed by a computer, cause the computer to implement the methods provided in the above embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0148] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0149] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

[0150] It should be understood that, in the embodiments of this specification, if the training and prediction processes of the AI ​​model are involved, the training and prediction processes of the AI ​​model shall comply with multiple legal and compliant principles, including legal data sources, compliant data content, compliant data governance, compliant training objectives and schemes, compliant training processes, compliant training environments and tools, and compliant ethical verification of training results, and shall meet the requirements of Article 5 of the Patent Law. Among them: Data source legitimacy: All datasets used for AI model training were obtained through legal means, covering three categories: publicly authorized data, data authorized by partners, and self-collected compliant data. Publicly authorized data comes from compliant data sources following open-source licenses such as Apache 2.0, with complete copyright attribution and authorization scope clearly marked, and no unauthorized open-source code or data reuse. Data authorized by partners has been subject to formal data usage agreements, clearly defining the scope, duration, and confidentiality obligations, and possessing a complete authorization chain. For self-collected data involving personal information, strict informed consent procedures have been followed, and anonymization processes (including but not limited to field masking, feature anonymization, and differential privacy technology applications) have been implemented to remove personally identifiable information, fully complying with the requirements of relevant laws and regulations such as the "Interim Measures for the Administration of Generative Artificial Intelligence Services" and the "Personal Information Protection Law."

[0151] Data content compliance: The AI ​​model's dataset undergoes multiple screenings and cleaning processes to remove all content that may violate social morality or harm public interests. It contains no obscene, pornographic, violent, discriminatory, or information that endangers national or public safety, nor does it involve the illegal acquisition or use of genetic resources. For data in sensitive fields (such as healthcare and finance), an additional privacy-preserving computation module (including federated learning and secure multi-party computation technologies) ensures that the data is "usable but not visible," avoiding compliance risks during the original data transmission process and ensuring that the data application scenarios and uses comply with public order and good morals and industry regulatory requirements.

[0152] Data governance norms: A complete data traceability system is established during the AI ​​model training process to automatically record the source, collection time, annotation process, cleaning rules, and permission allocation of training data, generating traceable compliance reports to ensure that the data is verifiable throughout its entire lifecycle. The dataset annotation process for AI models is completed by a professional human R&D team, clearly defining the proportion of human creative contributions and avoiding reliance on AI-generated data that has not undergone substantial human modification, thus meeting the examination requirements for "human main contributions" in AI patent applications.

[0153] Training objectives and plans are compliant: The AI ​​model training scheme and final output results do not violate any mandatory provisions of laws and administrative regulations, do not harm the public interest or the legitimate rights and interests of others, and do not pose any potential risks of being used for illegal activities, infringing on privacy, or disrupting public safety. They strictly adhere to the ethical principle of "intelligent for good".

[0154] Training process compliance: A closed-loop training framework is adopted to ensure compliance and controllability of the training process. The specific process is as follows: First, training samples are obtained through compliant data sources. After the aforementioned data cleaning and desensitization, they are input into the neural network model to generate preliminary training results. Second, an expert system is introduced to verify the preliminary results. Based on preset rules and human expert experience, the feasibility of the results is evaluated, and outputs that may pose ethical risks or compliance hazards are corrected (such as removing decision-making logic that violates public order and good morals, and adjusting model parameters that do not comply with safety regulations). Finally, the loss function weights are dynamically optimized based on expert system feedback to strengthen the model's learning of compliant results, avoid overfitting errors or non-compliant labels, and form a closed-loop control of "data input - model training - expert verification - parameter optimization - result feedback" to ensure that the entire training process complies with A5 ethical review requirements.

[0155] Training environment and tool compliance: AI model training is implemented using nationally licensed chips and a compliant training platform. All open-source frameworks and components used in the training process have obtained their corresponding licenses, and copyright statements and patent citation information are fully retained, with no instances of infringement or reuse. The training environment is built using virtual devices (containers / virtual machines) with fixed random seeds and initial parameter configurations to ensure the reproducibility of the training process. Furthermore, through access control and operation log recording, risks such as data leakage and parameter tampering during training are prevented, ensuring the security and compliance of the training process.

[0156] Training results ethical verification compliance: After the model is trained, it undergoes additional third-party ethical compliance assessment and algorithm filing review to verify that the model output does not violate social morality or harm public interests. For potentially sensitive scenarios (such as public services and intelligent decision-making), a special result verification mechanism is established to ensure that the model always complies with Article 5 of the Patent Law and relevant laws and regulations in practical applications.

[0157] In summary, the data and training process used in the AI ​​model of this specification strictly comply with the relevant provisions of Article 5 of the Patent Law and the Patent Examination Guidelines (2023 Edition), and there are no violations of laws, social ethics, public interests, or illegal use of genetic resources. It fully meets the compliance requirements for patent authorization.

Claims

1. An evaluation method, characterized in that, include: Obtain the symbol matrix and the effective angle domain channel matrix; The symbol matrix is ​​constructed based on the symbols sent by the user and received by the antenna array of the base station; the effective angle domain channel matrix is ​​constructed based on the angle domain channel vector from the user to the base station and the active indication parameter set for the user. The evaluation is performed based on a bilinear observation model; the evaluation object includes at least one of the following: whether the user is active, the user's signal strength, and the channel from the user to the base station. The bilinear observation model is constructed based on the symbol matrix, the effective angle domain channel matrix, and the received signal from the base station for the user.

2. The method as described in claim 1, characterized in that, The object to be evaluated includes the user's signal; therefore, based on the bilinear observation model, the object to be evaluated is evaluated, including: The expected value of the marginal posterior probability of the signal element in the symbol matrix in the bilinear observation model is calculated as an approximate estimate of the user's signal.

3. The method as described in claim 1 or 2, characterized in that, If the object to be evaluated includes whether the user is active, then, based on a bilinear observation model, the object to be evaluated is assessed, including: Based on the influence of the symbol matrix and the received signal on the effective angle domain channel matrix in the bilinear observation model, it is determined whether the user is an active user.

4. The method as described in claim 3, characterized in that, The influence results include: the observed mean of the channel elements in the effective angle domain channel matrix in the bilinear observation model; then, based on the influence results of the symbol matrix and the received signal on the effective angle domain channel matrix in the bilinear observation model, determining whether the user is an active user includes: Based on the observed mean, calculate the posterior probability weight of each channel element for which the channel element is nonzero; Based on the value of the posterior probability weight and the pre-set active user discrimination criterion based on the value of the posterior probability weight, it is determined whether the user is an active user.

5. The method as described in claim 3, characterized in that, The impact results include: the observation mean of the channel elements in the effective angle domain channel matrix in the bilinear observation model; the object to be evaluated includes the channel; therefore, the evaluation of the object to be evaluated based on the bilinear observation model also includes: Based on the posterior probability weight of the channel element, the Gaussian distribution variance when the channel element is non-zero, the noise variance of the observation mean, and the observation mean, the expected value of the marginal posterior probability of the channel element is calculated as an approximate estimation result of the channel element.

6. The method as described in claim 1, characterized in that, The method further includes: Obtain the spatial domain channel matrix; the spatial domain channel matrix is ​​constructed based on the spatial domain channel vector corresponding to the user. The spatial domain channel matrix is ​​converted into an angle domain channel matrix using the discrete Fourier transform matrix; the angle domain channel matrix contains the angle domain channel vector. The effective angle domain channel matrix is ​​constructed based on the angle domain channel matrix and the active indication parameter.

7. An evaluation device, characterized in that, include: The acquisition unit is used to acquire the symbol matrix and the effective angle domain channel matrix; The symbol matrix is ​​constructed based on the symbols sent by the user and received by the antenna array of the base station; the effective angle domain channel matrix is ​​constructed based on the angle domain channel vector from the user to the base station and the active indication parameter set for the user. The evaluation unit is used to evaluate the object to be evaluated based on a bilinear observation model. The object to be evaluated includes at least one of the following: whether the user is active, the user's signal strength, and the channel from the user to the base station. The bilinear observation model is constructed based on the symbol matrix, the effective angle domain channel matrix, and the received signal from the base station for the user.

8. A computing device, characterized in that, include: Memory and processor, among which, The memory is used to store computer programs; The processor, coupled to the memory, is configured to execute the computer program stored in the memory for performing the method described in any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, which, when executed by a computer, enables the implementation of the method described in any one of claims 1 to 6.

10. A computer program product, characterized in that, The computer program product stores instructions that, when executed by a computer, cause the computer to perform the method described in any one of claims 1 to 6.

Images