Parameter estimation method and device, electronic equipment, storage medium and program product

Abstract

The invention provides a parameter estimation method and device, electronic equipment, a storage medium and a program product, and relates to the field of channel measurement modeling. According to the method, multiple rounds of estimation can be carried out on the to-be-estimated parameters of the channel sub-paths in an iteration mode, and in each round of iteration, the method does not use a maximum likelihood estimation method to update the estimated values of the to-be-estimated parameters, but can use a likelihood function corresponding to the to-be-estimated parameters to generate a small amount of training data; and training the agent model of the to-be-estimated parameter by using the training data, so that the trained agent model is used for updating the estimated value of the to-be-estimated parameter, and calculation acceleration can be performed by using the agent model. Therefore, the agent model can be introduced to carry out channel sub-path multi-dimensional parameter estimation, and the calculation complexity can be reduced under the condition that the estimation precision is ensured.

Description

Technical Field

[0001] The present invention relates to the field of channel measurement modeling, and in particular to a parameter estimation method, device, electronic equipment, storage medium and program product. Background Art

[0002] The estimation of multidimensional parameters of channel subpaths is a key issue in the channel measurement and modeling process, and has important applications in practical projects such as radar and indoor wireless positioning of communication systems. In the estimation of multidimensional parameters of channel subpaths, it is generally necessary to estimate the signal amplitude, signal delay, two-dimensional departure angle, and two-dimensional arrival angle of the channel subpaths. The current mainstream multidimensional parameter estimation methods are divided into four categories: spectral estimation methods, methods based on subspace decomposition, methods based on maximum likelihood, and data-driven methods. However, related multidimensional parameter estimation methods have the defects of low accuracy or high computational complexity. Summary of the invention

[0003] The purpose of the present invention is to provide a parameter estimation method, device, electronic device, storage medium and program product, which can introduce a proxy model to perform multi-dimensional parameter estimation of channel subpaths and reduce the computational complexity while ensuring the estimation accuracy.

[0004] In order to solve the above technical problems, the present invention provides a parameter estimation method, comprising:

[0005] When entering this round of iteration, the likelihood function corresponding to the parameter to be estimated in the multidimensional parameters is determined according to the latest estimated value corresponding to the multidimensional parameters of the channel subpath; the multidimensional parameters include the signal amplitude and the parameter to be estimated, and the parameter to be estimated includes the signal delay, the two-dimensional departure angle and the two-dimensional arrival angle;

[0006] Sampling in the parameter space corresponding to the parameter to be estimated to obtain a parameter value, and inputting the parameter value into the likelihood function corresponding to the parameter to be estimated to obtain a response value;

[0007] Training a proxy model of the parameter to be estimated using the parameter value and the response value, and updating an estimated value of the parameter to be estimated using the trained proxy model;

[0008] Updating the estimated value of the signal amplitude using the latest estimated value corresponding to the parameter to be estimated;

[0009] Whether to end the iteration is determined according to the preset iteration end condition, and when it is determined not to exit the iteration, the next round of iteration is entered, or when it is determined to exit the iteration, the latest corresponding estimated value of the multidimensional parameter is output.

[0010] Optionally, the step of training a proxy model of the parameter to be estimated by using the parameter value and the response value, and updating the estimated value of the parameter to be estimated by using the trained proxy model, comprises:

[0011] Based on a Gaussian process regression method, the proxy model is established and trained using the parameter values ​​and the response values;

[0012] In the trained proxy model, the maximum value of the parameter to be estimated is searched based on the gradient descent algorithm, and the maximum value of the parameter to be estimated is used as the estimated value for updating the parameter to be estimated.

[0013] Optionally, sampling in a parameter space corresponding to the parameter to be estimated to obtain a parameter value includes:

[0014] The parameter value is obtained by sampling in the parameter space corresponding to the parameter to be estimated using the Latin hypercube method.

[0015] Optionally, before determining whether to end the iteration according to a preset iteration end condition, the method further includes:

[0016] Determining whether the multidimensional parameters of each of the channel subpaths have completed the update of estimated values ​​in this round of iteration;

[0017] If yes, then enter the step of determining whether to end the iteration according to the preset iteration end condition;

[0018] If not, for the next channel sub-path for which the estimated value update has not been completed, the step of determining the likelihood function corresponding to the parameter to be estimated in the multi-dimensional parameters according to the latest estimated value corresponding to the multi-dimensional parameters of the channel sub-path is entered.

[0019] Optionally, determining a likelihood function corresponding to a parameter to be estimated in the multidimensional parameter according to the latest estimated value corresponding to the multidimensional parameter of the channel subpath includes:

[0020] Determining the expected signal value corresponding to the channel subpath in this iteration according to the latest corresponding estimated value of the multidimensional parameter of the channel subpath;

[0021] The likelihood function corresponding to the parameter to be estimated is determined by using the expected signal value and the latest estimated value corresponding to the parameter to be estimated.

[0022] Optionally, the determining a likelihood function corresponding to the parameter to be estimated by using the expected signal value and the latest estimated value corresponding to the parameter to be estimated includes:

[0023] Determine a likelihood function corresponding to the signal delay by using the expected signal value, the latest estimated value corresponding to the two-dimensional departure angle, and the latest estimated value corresponding to the two-dimensional arrival angle;

[0024] When the estimated value of the signal delay is updated in this round of iteration, the likelihood function corresponding to the two-dimensional departure angle is determined using the expected signal value, the latest estimated value corresponding to the signal delay, and the latest estimated value corresponding to the two-dimensional arrival angle;

[0025] When the estimated value of the two-dimensional departure angle is updated in this round of iteration, the likelihood function corresponding to the two-dimensional arrival angle is determined using the expected signal value, the latest estimated value corresponding to the signal delay and the latest estimated value corresponding to the two-dimensional departure angle.

[0026] The present invention also provides a parameter estimation device, comprising:

[0027] A likelihood function determination module, used to determine, when entering this round of iteration, the likelihood function corresponding to the parameter to be estimated in the multidimensional parameters according to the latest corresponding estimated value of the multidimensional parameters of the channel subpath; the multidimensional parameters include the signal amplitude and the parameter to be estimated, and the parameter to be estimated includes the signal delay, the two-dimensional departure angle and the two-dimensional arrival angle;

[0028] A response value determination module, used for sampling in the parameter space corresponding to the parameter to be estimated to obtain a parameter value, and inputting the parameter value into the likelihood function corresponding to the parameter to be estimated to obtain a response value;

[0029] A proxy model processing module, used to train the proxy model of the parameter to be estimated using the parameter value and the response value, and update the estimated value of the parameter to be estimated using the trained proxy model;

[0030] A signal amplitude updating module, used to update the estimated value of the signal amplitude using the latest estimated value corresponding to the parameter to be estimated;

[0031] The iteration control module is used to determine whether to end the iteration according to the preset iteration end condition, and enter the next round of iteration when it is determined not to exit the iteration, or output the latest corresponding estimated value of the multidimensional parameter when it is determined to exit the iteration.

[0032] The present invention also provides an electronic device, comprising:

[0033] Memory for storing computer programs;

[0034] A processor is used to implement the parameter estimation method as described above when executing the computer program.

[0035] The present invention also provides a computer program product, comprising a computer program or instructions, wherein when the computer program or instructions are executed by a processor, the parameter estimation method as described above is implemented.

[0036] The present invention also provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are loaded and executed by a processor, the parameter estimation method as described above is implemented.

[0037] The present invention provides a parameter estimation method, comprising: when entering this round of iteration, determining the likelihood function corresponding to the parameter to be estimated in the multidimensional parameters according to the latest corresponding estimated value of the multidimensional parameters of the channel subpath; the multidimensional parameters include the signal amplitude and the parameter to be estimated, and the parameter to be estimated includes the signal delay, the two-dimensional departure angle and the two-dimensional arrival angle; sampling in the parameter space corresponding to the parameter to be estimated to obtain the parameter value, and inputting the parameter value into the likelihood function corresponding to the parameter to be estimated to obtain the response value; training the proxy model of the parameter to be estimated by using the parameter value and the response value, and updating the estimated value of the parameter to be estimated by using the trained proxy model; updating the estimated value of the signal amplitude by using the latest corresponding estimated value of the parameter to be estimated; judging whether to end the iteration according to a preset iteration end condition, and entering the next round of iteration when it is determined that the iteration is not to be exited, or outputting the latest corresponding estimated value of the multidimensional parameter when it is determined that the iteration is to be exited.

[0038] The beneficial effect of the present invention is that the present invention can perform multiple rounds of estimation of the multidimensional parameters of the channel subpath in an iterative manner, and can use a proxy model to accelerate the maximum likelihood estimation stage of each round of iteration to ensure both estimation accuracy and estimation efficiency. Specifically, when entering this round of iteration, the present invention can determine the likelihood function corresponding to the parameter to be estimated in the multidimensional parameters according to the latest corresponding estimated value of the multidimensional parameters of the channel subpath; the multidimensional parameters include signal amplitude and the parameter to be estimated, and the parameter to be estimated includes signal delay, two-dimensional departure angle and two-dimensional arrival angle. Subsequently, sampling is performed in the parameter space corresponding to the parameter to be estimated to obtain the parameter value, and the parameter value is input into the likelihood function corresponding to the parameter to be estimated to obtain the response value. Subsequently, the parameter value and the response value can be used to train the proxy model of the parameter to be estimated, and the trained proxy model can be used to update the estimated value of the parameter to be estimated. That is to say, in simple terms, the likelihood function in the present invention is not used for maximum likelihood estimation, but is used to generate a small amount of training data (i.e., parameter values ​​and response values); the present invention does not update the estimated value of the parameter to be estimated based on the maximum likelihood estimation, but can use the training data to train the proxy model of the parameter to be estimated, and use the trained proxy model to update the estimated value of the parameter to be estimated, that is, the proxy model can be used to approximate the effect of maximum likelihood estimation. In this way, the present embodiment can use the proxy model to reduce the computational complexity while ensuring the estimation accuracy. Subsequently, the present invention can use the latest corresponding estimated value of the parameter to be estimated to update the estimated value of the signal amplitude, and determine whether to end the iteration according to the preset iteration end condition, so as to enter the next round of iteration when it is determined that the iteration is not to be exited, and output the latest corresponding estimated value of the multidimensional parameter when it is determined to exit the iteration, so as to obtain a reliable estimated value after multiple rounds of iteration. The present invention also provides a parameter estimation device, an electronic device, a computer-readable storage medium, and a computer program product, which have the above-mentioned beneficial effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0040] Figure 1 A schematic diagram of a signal model provided by an embodiment of the present invention;

[0041] Figure 2 A flow chart of a parameter estimation method provided by an embodiment of the present invention;

[0042] Figure 3 A flowchart of another parameter estimation method provided by an embodiment of the present invention;

[0043] Figure 4 A comparison diagram of the response effects of the likelihood function of the delay parameter and different proxy functions of the delay parameter provided in the embodiment of the present invention;

[0044] Figure 5 A comparison diagram of the response effects of the likelihood function of the two-dimensional departure angle parameter and different proxy functions of the two-dimensional departure angle parameter provided by an embodiment of the present invention;

[0045] Figure 6 A schematic diagram of a simulation result provided by an embodiment of the present invention;

[0046] Figure 7 A structural block diagram of a parameter estimation device provided by an embodiment of the present invention;

[0047] Figure 8 The present invention provides a structural block diagram of an electronic device. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] The estimation of multidimensional parameters of channel subpaths is a key issue in the channel measurement and modeling process, and has important applications in practical projects such as radar and indoor wireless positioning of communication systems. For ease of understanding, the following first briefly introduces the relevant concepts in the estimation of multidimensional parameters of channel subpaths.

[0050] Please refer to Figure 1 , Figure 1 A schematic diagram of a signal model provided in an embodiment of the present invention shows a hypothetical broadband MIMO array (Multiple-Input Multiple-Output). The array is a two-dimensional planar array, which is divided into a transmitting array and a receiving array. The transmitting array includes M transmitting array elements, and the receiving array includes N receiving array elements. Assuming that the k far-field signals (corresponding to k channel subpaths) received by the array have different two-dimensional departure angles, two-dimensional arrival angles, delays, and amplitudes, then and The frequencies are The transmit steering vector and receive steering vector on can be expressed as:

[0051] ;

[0052] In the above formula, r represents the receiving end, t represents the transmitting end, , represents the position coordinates of the array element at the nth receiving end, represents the direction cosine vector of the kth far-field signal at the receiving end, represents the pitch angle, represents the azimuth, T represents the transpose, represents the position coordinates of the array element at the nth transmitter, Represents the direction cosine vector of the kth far-field signal at the transmitter.

[0053] The array receiving signal at the qth frequency point is expressed as:

[0054] ;

[0055] Then, the array receiving signal can be expressed as:

[0056] ;

[0057] in, represents the multidimensional parameters to be estimated for the k-th channel subpath, which includes 6-dimensional parameters, represents the signal amplitude of the kth subpath, represents the signal delay of the kth subpath, Together they represent the two-dimensional departure angle of the kth subpath, Together they represent the two-dimensional arrival angle of the kth subpath. represents the noise term; represents the variance of the noise.

[0058] The channel transfer function of the kth subpath at Q frequency points is expressed as:

[0059] ;

[0060] in, The function represents the channel transfer function in extracting the signal amplitude The remaining part after the semicolon, in the parameters of this function, the parameters before the semicolon represent the parameters to be estimated, and the parameters after the semicolon represent fixed values. Represents the time delay, two-dimensional departure angle, and two-dimensional arrival angle parameters in the k-th channel subpath excluding the amplitude parameter. Represents the signal frequency, which is a fixed value in this article.

[0061] Based on the above introduction, among the related technologies of channel subpath multidimensional parameter estimation, the current mainstream multidimensional parameter estimation methods are divided into four categories: spectral estimation method, subspace decomposition-based method, maximum likelihood-based method, and data-driven method. However, the estimation resolution of the spectral estimation method is very low, and the estimation performance is very poor under low signal-to-noise ratio. The method based on subspace decomposition has high computational complexity, harsh application conditions, and few applicable scenarios. The method based on maximum likelihood has high computational complexity and cannot perform real-time estimation. The data-driven method has poor generalization performance and ignores the joint estimation of multidimensional parameters. In short, the related multidimensional parameter estimation methods have the defects of low accuracy or high computational complexity.

[0062] In view of this, in order to solve the technical problem of how to improve the estimation efficiency while ensuring the accuracy of channel sub-path multi-dimensional parameter estimation, the present invention can provide a parameter estimation method, introduce a proxy model to perform channel sub-path multi-dimensional parameter estimation, and use the proxy model to replace the maximum likelihood estimation, thereby reducing the computational complexity while ensuring the estimation accuracy.

[0063] For easier understanding, please refer to Figure 2 , Figure 2 A flow chart of a parameter estimation method provided by an embodiment of the present invention. The method may include:

[0064] S101. When entering this round of iteration, determine the likelihood function corresponding to the parameter to be estimated in the multidimensional parameters according to the latest corresponding estimated value of the multidimensional parameters of the channel subpath; the multidimensional parameters include signal amplitude and the parameter to be estimated, and the parameter to be estimated includes signal delay, two-dimensional departure angle and two-dimensional arrival angle.

[0065] In this embodiment, the multidimensional parameters of the channel subpath may include signal amplitude, signal delay, two-dimensional departure angle, and two-dimensional arrival angle. These multidimensional parameters may be updated with estimated values ​​through multiple rounds of iterations. In each round of iteration, the signal amplitude can generally be obtained in one step through a closed-form solution; while the signal delay, two-dimensional departure angle, and two-dimensional arrival angle generally require the use of a maximum likelihood estimation method for parameter estimation in related technologies, which has a high computational complexity. Therefore, in this embodiment, the signal delay, two-dimensional departure angle, and two-dimensional arrival angle may be collectively referred to as parameters to be estimated, and the estimation method of these parameters to be estimated is particularly improved.

[0066] Furthermore, this step first needs to determine the likelihood function corresponding to the parameter to be estimated in the multidimensional parameters according to the latest corresponding estimated value of the multidimensional parameters of the channel subpath (i.e., the estimated value updated in the most recent parameter estimation iteration of the multidimensional parameters). Specifically, the expected signal value corresponding to the channel subpath in this iteration can be determined based on the latest corresponding estimated value of the multidimensional parameters of the channel subpath. Subsequently, the likelihood function corresponding to the parameter to be estimated can be determined using the expected signal value and the latest corresponding estimated value of the parameter to be estimated.

[0067] Based on this, determining the likelihood function corresponding to the parameter to be estimated in the multidimensional parameters according to the latest estimated value corresponding to the multidimensional parameters of the channel subpath may include:

[0068] Step 11: Determine the expected signal value corresponding to the channel subpath in this iteration according to the latest corresponding estimated value of the multidimensional parameter of the channel subpath.

[0069] Specifically, the expected signal value corresponding to the channel subpath in this round of iteration can be expressed as:

[0070] ;

[0071] in, represents the expected signal value corresponding to the kth subpath in the i-th round of parameter estimation iteration, represents the original array receiving signal, represents the signal amplitude estimate corresponding to the k'th subpath in the i-th round of parameter estimation iteration, It represents the signal delay estimation value, two-dimensional departure angle estimation value and two-dimensional arrival angle estimation value corresponding to the k'th sub-path in the i-th round of parameter estimation iteration.

[0072] Step 12: Determine the likelihood function corresponding to the parameter to be estimated using the expected signal value and the latest estimated value corresponding to the parameter to be estimated.

[0073] Taking the delay parameter as an example, the likelihood function of the delay parameter can be expressed as:

[0074] ;

[0075] Among them, vec means vectorization, represents the signal delay corresponding to the k-th subpath in the i+1th round of parameter estimation iteration, which is a variable in the likelihood function and a parameter to be estimated; represents the 2D departure angle and 2D arrival angle corresponding to the kth subpath in the i-th round of parameter estimation iteration, which are fixed values ​​in the likelihood function, and The values ​​of are the estimated values ​​of the two-dimensional departure angle and the two-dimensional arrival angle determined in the i-th round of parameter estimation iteration. express Conjugate transpose of a function.

[0076] It can be seen that this embodiment can determine the likelihood function of each parameter to be estimated in this round of iteration according to the estimated value of the multidimensional parameter in the previous round of parameter estimation. Therefore, it can be understood that when performing the first round of iteration, it is necessary to initialize the estimated value of each multidimensional parameter to construct the likelihood function of the first round of iteration based on the initial estimated value.

[0077] Based on this, before determining the likelihood function corresponding to the parameter to be estimated in the multidimensional parameters according to the latest estimated value corresponding to the multidimensional parameters of the channel subpath, the following may also be included:

[0078] Step 21: When it is determined that this round of iteration is the first round of iteration, the estimated values ​​of the multidimensional parameters are initialized.

[0079] It should be noted that the present embodiment does not limit the specific initialization method, which can be set according to actual application requirements, for example, random initialization or all-zero initialization.

[0080] S102 . Sampling is performed in a parameter space corresponding to the parameter to be estimated to obtain a parameter value, and the parameter value is input into a likelihood function corresponding to the parameter to be estimated to obtain a response value.

[0081] It is worth noting that after the likelihood function is obtained in the previous step, this step does not use the likelihood function for maximum likelihood estimation, but can use the likelihood function to generate training data. Specifically, this step can sample a small number of parameter values ​​in the parameter space corresponding to the parameter to be estimated, and input these parameter values ​​into the likelihood function corresponding to the parameter to be estimated to obtain the response value, thereby correspondingly forming training data with the parameter value and the response value.

[0082] It should be noted that the present embodiment does not limit the number of sampling of parameter values, which can be set according to actual application requirements. The present embodiment does not limit the specific sampling method, which can also be selected according to actual requirements. For example, the Latin hypercube method can be used to sample in the parameter space corresponding to the parameter to be estimated to obtain the parameter value.

[0083] S103: Train a proxy model of the parameter to be estimated using the parameter value and the response value, and update the estimated value of the parameter to be estimated using the trained proxy model.

[0084] It is also worth noting that in this step, the proxy model of the parameter to be estimated can be trained using the parameter values ​​and the response values ​​to optimize the model parameters of the proxy model, and then the trained proxy model can be used to update the estimated value of the parameter to be estimated in this round of parameter estimation iteration. Considering that the proxy model is trained using the actual parameter values ​​and response values, the present embodiment can use the proxy model to reasonably replace the maximum likelihood function, and can achieve an accuracy close to the maximum likelihood estimation through the proxy model. At the same time, considering that in each round of parameter estimation iteration, the present embodiment only needs to use a small number of parameter values ​​in the parameter space of the parameter to be estimated to train the proxy model, and the proxy model can be used for parameter estimation without the need for maximum likelihood estimation based on the complete parameter space of the parameter to be estimated. Therefore, the present embodiment can significantly reduce the amount of calculation in each round of iteration, thereby improving the efficiency of parameter estimation.

[0085] It should also be pointed out that the proxy model in this embodiment needs to be dynamically trained to ensure that it can closely match the actual channel environment.

[0086] It should be noted that the present embodiment does not limit the specific type of the proxy model, and can be set according to actual application requirements, such as a Gaussian process regression model, a support vector machine, a neural network, etc. The present embodiment also does not limit the specific method of training the proxy model, and can be set according to actual application requirements. In order to achieve a better estimation effect, the present embodiment can construct and train the proxy model based on the Gaussian process regression method (GPR, Gaussian Process Regression), and can search for the maximum value of the parameter to be estimated based on the gradient descent algorithm in the trained proxy model, so as to use the maximum value of the parameter to be estimated as the estimated value of the parameter to be estimated to update, so as to use the model gradient descent optimization to replace the maximum likelihood estimation.

[0087] Based on this, training a proxy model of the parameter to be estimated using the parameter value and the response value, and updating the estimated value of the parameter to be estimated using the trained proxy model may include:

[0088] Step 31: Based on the Gaussian process regression method, the proxy model is established and trained using the parameter values ​​and the response values.

[0089] In this step, you can first build a proxy model based on the Gaussian process regression method, such as selecting a suitable kernel function (Covariance Function) as the core of the proxy model. Common kernel functions include RBF kernel (radial basis function), Matern kernel, etc. You can refer to the relevant technologies of Gaussian process regression. Then, you can use parameter values ​​and response values ​​to train the proxy model to optimize the hyperparameters in the proxy model (such as the parameters of the kernel function), so that the proxy model fits the training data (i.e. parameter values ​​and response values).

[0090] Step 32: In the trained proxy model, the maximum value of the parameter to be estimated is searched based on the gradient descent algorithm, and the maximum value of the parameter to be estimated is used as the estimated value for updating the parameter to be estimated.

[0091] In this step, the trained proxy model will be used to search for the maximum value of the parameter to be estimated, so that the maximum value is used as the estimated value of the parameter to be estimated to be updated in this round of parameter estimation iteration. In the related art, this search is usually performed based on the grid search algorithm, and the algorithm searches based on the idea of ​​exhaustive method, which has the defects of large search volume and low search efficiency. To this end, the present invention can use the gradient descent algorithm to replace the grid search algorithm. The core principle of the gradient descent algorithm is to use gradient information to guide the search direction. Mathematically, the gradient can be obtained by differentiating the function. For a multivariate function, the gradient is a vector, each component of which corresponds to the partial derivative of the function with respect to each variable. By calculating the gradient of the current point, the direction in which the function grows (or decreases) fastest at this point can be determined, thereby guiding the next search. In this way, the maximum value of the parameter to be estimated can be searched based on the gradient descent algorithm in the trained proxy model, which can avoid exhaustive search, thereby improving the search efficiency. Specifically, the gradient descent algorithm is an iterative process, and the final parameter to be estimated is obtained by continuous iteration, which specifically includes the following steps:

[0092] Step 31: Calculate the gradient of the proxy model for the parameter to be estimated in the current gradient descent iteration. The gradient can be expressed as:

[0093] ;

[0094] in, It represents the estimated value of the parameter to be estimated for the kth subpath in the i+1th round of parameter estimation iteration and the stepth round of gradient descent iteration. Step represents the number of gradient descent iterations. It should be noted that It represents a parameter to be estimated among the signal delay, two-dimensional departure angle, and two-dimensional arrival angle. If the current gradient descent iteration round is the first round, The value of needs to be initialized. Represents a trained proxy model.

[0095] It should be noted that this embodiment does not limit how to calculate the gradient, and reference may be made to related technologies for gradient calculation.

[0096] Step 32: Update the parameters to be estimated for the next gradient descent iteration according to the gradient. The update process can be expressed as:

[0097]

[0098] in, Indicates the estimated value of the parameter to be estimated for the k-th subpath in the i+1th round of parameter estimation iteration and the step+1th round of gradient descent iteration; Represents the learning rate, which is a hyperparameter.

[0099] Step 33: Determine whether the preset gradient descent convergence condition is met. If so, As the maximum value of the parameter to be estimated, and use the maximum value as the estimated value updated in the i+1th round of parameter estimation iteration; if it is not satisfied, enter the next round of gradient descent iteration.

[0100] It should be noted that the present embodiment does not limit the specific gradient descent convergence condition. For example, it can be used to determine whether the gradient descent iteration round reaches a preset round upper limit value. If it reaches, it is determined that the condition is satisfied, and if it does not reach, it is determined that the condition is not satisfied. For another example, it can also be used to determine whether the gradient value is less than a preset threshold value. If it is less than, it is determined that the condition is satisfied, and if it is not less than, it is determined that the condition is not satisfied. The above two gradient descent convergence conditions can be selected according to actual application requirements.

[0101] Further, in order to effectively perform multi-dimensional parameter joint estimation of the signal instead of single-dimensional parameter estimation, the present embodiment can also estimate each parameter to be estimated in sequence, and after completing the update of the estimated value of one parameter to be estimated, the latest estimated value of the parameter to be estimated can be used to estimate the next parameter to be estimated. It should be noted that the present embodiment does not limit the specific estimation order of the parameters to be estimated, and can be set according to actual application requirements. For example, the signal delay can be estimated first; after completing the signal delay estimation, the two-dimensional departure angle can be estimated based on the latest estimated value of the signal delay in this round of estimation; after completing the two-dimensional departure angle estimation, the two-dimensional arrival angle can be estimated based on the latest estimated values ​​of the signal delay and the two-dimensional departure angle in this round of estimation. Furthermore, under this order, each likelihood function also has a corresponding generation order, that is, the likelihood function of the signal delay is generated first; after completing the signal delay estimation, the likelihood function of the two-dimensional departure angle is generated based on the latest estimated value of the signal delay in this round of estimation; after completing the two-dimensional departure angle estimation, the likelihood function of the two-dimensional arrival angle is generated based on the latest estimated value of the signal delay and the two-dimensional departure angle in this round of estimation.

[0102] Based on this, determining the likelihood function corresponding to the parameter to be estimated by using the expected signal value and the latest estimated value corresponding to the parameter to be estimated may include:

[0103] Step 41: Determine a likelihood function corresponding to a signal delay using an expected signal value, a latest estimated value corresponding to a two-dimensional departure angle, and a latest estimated value corresponding to a two-dimensional arrival angle;

[0104] Step 42: when the estimated value of the signal delay is updated in this iteration, the likelihood function corresponding to the two-dimensional departure angle is determined using the expected signal value, the latest estimated value corresponding to the signal delay, and the latest estimated value corresponding to the two-dimensional arrival angle;

[0105] Step 43: When the estimated value of the two-dimensional departure angle is updated in this iteration, the likelihood function corresponding to the two-dimensional arrival angle is determined using the expected signal value, the latest estimated value corresponding to the signal delay and the latest estimated value corresponding to the two-dimensional departure angle.

[0106] The estimation process of each parameter to be estimated is introduced below based on specific formulas.

[0107] 1. Generate a likelihood function of signal delay, train a proxy model of signal delay, and update the estimated value of signal delay.

[0108] The likelihood function of signal delay is expressed as:

[0109] ;

[0110] in, represents the signal delay corresponding to the k-th subpath in the i+1th round of parameter estimation iteration, which is a variable in the likelihood function and a parameter to be estimated; represents the 2D departure angle and 2D arrival angle corresponding to the kth subpath in the i-th round of parameter estimation iteration, which are fixed values ​​in the likelihood function, and The value of is the estimated value of the two-dimensional departure angle and the two-dimensional arrival angle determined in the i-th round of parameter estimation iteration. Substituting the parameter value obtained by sampling signal delay into the function formula can obtain the corresponding response value.

[0111] The proxy model of signal delay is expressed as:

[0112] ;

[0113] in, Represents a proxy model.

[0114] 2. After completing the update of the estimated value of the signal delay, generate the likelihood function of the two-dimensional departure angle based on the latest estimated value of the signal delay, train the proxy model of the two-dimensional departure angle, and update the estimated value of the two-dimensional departure angle.

[0115] The likelihood function of the two-dimensional departure angle is expressed as:

[0116] ;

[0117] in, It represents the two-dimensional departure angle corresponding to the k-th sub-path in the i+1th round of parameter estimation iteration, which is a variable in the likelihood function and a parameter to be estimated; represents the signal delay corresponding to the kth subpath in the i+1th parameter estimation iteration and the two-dimensional arrival angle corresponding to the subpath in the i-th parameter estimation iteration, which are fixed values ​​in the likelihood function, and Updated to the latest estimated value. Substitute the parameter value obtained by sampling the two-dimensional departure angle into the function formula to get the corresponding response value.

[0118] The proxy model of the two-dimensional departure angle is expressed as:

[0119] .

[0120] 3. After completing the update of the estimated value of the two-dimensional departure angle, generate the likelihood function of the two-dimensional arrival angle based on the signal delay and the latest estimated value of the two-dimensional departure angle, train the proxy model of the two-dimensional arrival angle, and update the estimated value of the two-dimensional arrival angle.

[0121] The likelihood function of the two-dimensional arrival angle is expressed as:

[0122] ;

[0123] in, It represents the two-dimensional arrival angle corresponding to the k-th subpath in the i+1th round of parameter estimation iteration, which is a variable in the likelihood function; represents the signal delay and two-dimensional departure angle corresponding to the k-th subpath in the i+1th round of parameter estimation iteration, which are fixed values ​​in the likelihood function, and All have been updated to the latest estimated values. Substituting the parameter values ​​obtained by sampling the two-dimensional arrival angle into the function formula can obtain the corresponding response value.

[0124] The proxy model for the two-dimensional arrival angle is expressed as:

[0125] .

[0126] S104: Update the estimated value of the signal amplitude using the latest estimated value corresponding to the parameter to be estimated.

[0127] In this step, the estimated value of the signal amplitude can be obtained in one step based on the latest estimated values ​​of the signal delay, the two-dimensional departure angle, and the two-dimensional arrival angle, which is expressed as:

[0128] ;

[0129] in, represents the estimated value of the signal amplitude corresponding to the kth subpath in the i+1th iteration. The S() function here is slightly different from the S() function above. The difference is reflected in the S() function here. are fixed values, i.e. It represents the estimated values ​​of the delay parameter, the two-dimensional departure angle, and the two-dimensional arrival angle determined in the i+1th round of parameter estimation iteration.

[0130] S105, judging whether to end the iteration according to the preset iteration end condition, and entering the next round of iteration when it is determined not to exit the iteration, or outputting the latest corresponding estimated value of the multidimensional parameter when it is determined to exit the iteration.

[0131] In this step, whether to end the iteration can be determined according to the preset iteration end condition. This embodiment does not limit the specific preset iteration end condition, which can be set according to the actual application requirements. For example, the preset iteration end condition can determine whether the multidimensional parameters of the channel sub-path converge (that is, whether the difference between the estimated value of this round of iteration and the estimated value of the previous round of iteration is less than the preset threshold value). If it is determined that the multidimensional parameter has converged, it can be determined to exit the iteration, otherwise it is necessary to enter the next round of iteration. For another example, the preset iteration end condition can determine whether the executed iteration rounds have reached a preset upper limit value, and the executed iteration rounds are the total number of executed iteration rounds. If the executed iteration rounds reach the preset upper limit value, it can be determined to exit the iteration, otherwise it is necessary to enter the next round of iteration. It should be noted that the above two methods can be selected arbitrarily or used in combination.

[0132] Based on this, judging whether to end the iteration according to the preset iteration end condition may include:

[0133] Step 51: Determine whether the number of executed iterations has reached a preset upper limit; if the number of executed iterations has reached the preset upper limit, proceed to step 52; if the number of executed iterations has not reached the preset upper limit, proceed to step 53;

[0134] Step 52: Determine whether to exit iteration;

[0135] Step 53: Determine whether to exit iteration;

[0136] Or, judging whether to end the iteration according to a preset iteration end condition may include:

[0137] Step 61: determine whether the difference between the estimated value of the parameter to be estimated determined in this round of iteration and the estimated value determined in the previous round of iteration is less than a preset threshold; if the difference is less than the preset threshold, proceed to step 62; if the difference is not less than the preset threshold, proceed to step 63;

[0138] Step 62: Determine whether to exit iteration;

[0139] Step 63: Determine whether to exit iteration.

[0140] Of course, considering that each round of iteration requires updating the estimated values ​​of the multidimensional parameters of each channel sub-path, before determining whether to end the iteration, it is also possible to determine whether the multidimensional parameters of each channel sub-path have all completed the estimated value update in this round of iteration. If they have all been updated, the iteration exit judgment step can be entered. Otherwise, it is necessary to continue updating the estimated values ​​of the channel sub-paths that have not been updated.

[0141] Based on this, before judging whether to end the iteration according to the preset iteration end condition, the following is also included:

[0142] Step 71: determine whether the multidimensional parameters of each of the channel subpaths have completed the update of estimated values ​​in this round of iteration; if so, proceed to step 72; if not, proceed to step 73;

[0143] Step 72: Entering the step of determining whether to end iteration according to a preset iteration end condition;

[0144] Step 73: For the next channel sub-path for which the estimated value update has not been completed, enter the step of determining the likelihood function corresponding to the to-be-estimated parameter in the multi-dimensional parameters according to the latest estimated value corresponding to the multi-dimensional parameters of the channel sub-path.

[0145] Finally, to fully understand the parameter estimation method provided in this embodiment, please refer to Figure 3 , Figure 3 This is a flow chart of another parameter estimation method provided by an embodiment of the present invention, which fully demonstrates the execution process of the method.

[0146] Based on the above embodiments, the present invention can perform multiple rounds of estimation on the multidimensional parameters of the channel subpath in an iterative manner, and can use a proxy model to accelerate the maximum likelihood estimation stage of each round of iteration to ensure both estimation accuracy and estimation efficiency. Specifically, when entering this round of iteration, the present invention can determine the likelihood function corresponding to the parameter to be estimated in the multidimensional parameters according to the latest corresponding estimated value of the multidimensional parameters of the channel subpath; the multidimensional parameters include signal amplitude and the parameter to be estimated, and the parameter to be estimated includes signal delay, two-dimensional departure angle and two-dimensional arrival angle. Subsequently, sampling is performed in the parameter space corresponding to the parameter to be estimated to obtain the parameter value, and the parameter value is input into the likelihood function corresponding to the parameter to be estimated to obtain the response value. Subsequently, the parameter value and the response value can be used to train the proxy model of the parameter to be estimated, and the trained proxy model can be used to update the estimated value of the parameter to be estimated. In short, the likelihood function in the present invention is not used for maximum likelihood estimation, but for generating a small amount of training data (i.e., parameter values ​​and response values); the present invention does not update the estimated value of the parameter to be estimated based on the maximum likelihood estimation, but can use the training data to train the proxy model of the parameter to be estimated, and use the trained proxy model to update the estimated value of the parameter to be estimated, that is, the proxy model can be used to approximate the effect of maximum likelihood estimation. In this way, the present embodiment can use the proxy model to reduce the computational complexity while ensuring the estimation accuracy. Subsequently, the present invention can use the latest corresponding estimated value of the parameter to be estimated to update the estimated value of the signal amplitude, and determine whether to end the iteration according to the preset iteration end condition, so as to enter the next round of iteration when it is determined that the iteration is not to be exited, and output the latest corresponding estimated value of the multidimensional parameter when it is determined to exit the iteration, so as to obtain a reliable estimated value after multiple rounds of iterations.

[0147] Based on the above embodiments, the parameter estimation effect achieved by this method is introduced below. Figure 4 , Figure 5 . Figure 4 The figure shows the comparison between the likelihood function of the delay parameter of the first sub-path and the likelihood functions reconstructed by different proxy functions. The upper left corner shows the response generated by the likelihood function of the delay parameter, the upper right corner shows the response generated by the likelihood function reconstructed by the Gaussian process regression proxy model (GPR), the lower left corner shows the response generated by the likelihood function reconstructed by the support vector machine (SVR, Support Vector Regression), and the lower right corner shows the response generated by the likelihood function reconstructed by the neural network (NN, Neural Network). Figure 5The figure shows the comparison between the likelihood function of the two-dimensional departure angle parameter of the first sub-path and the likelihood function reconstructed by the surrogate function. The upper left corner shows the response generated by the likelihood function of the two-dimensional departure angle, the upper right corner shows the response generated by the likelihood function reconstructed by the Gaussian process regression surrogate model (GPR), the lower left corner shows the response generated by the likelihood function reconstructed by the support vector machine (SVR), and the lower right corner shows the response generated by the likelihood function reconstructed by the neural network (NN). Figure 4 and Figure 5 It can be seen that the Gaussian process regression proxy model (GPR) can better approximate the likelihood function of the delay and angle parameters.

[0148] This embodiment can illustrate that the estimation accuracy of this method (ML SAGE) is higher than that of the traditional SAGE algorithm and the maximum likelihood method (MLE) with low computational complexity through the following simulation experiments:

[0149] (A) Experimental setup:

[0150] This embodiment uses a 5.4 GHz carrier frequency and an 80 MHz bandwidth OFDM signal as the transmission signal. The number of subcarriers is 256. Both the transmission array and the receiving array are The uniform planar array has an element spacing of half a wavelength.

[0151] (B) Statistical performance comparison:

[0152] The root mean square error (RMSE) can be used to evaluate the statistical performance. In each signal noise, 100 Monte Carlo simulations were performed to eliminate randomness. The calculation formula of RMSE is as follows:

[0153]

[0154] in It is in The estimated parameters in this simulation. Please refer to Figure 6 , where the derivation basis of CRB and simulation results are shown in Figure 6 , where Delay is the estimated mean square error of the delay, and AoD is the estimated azimuth departure angle and elevation departure angle ( ), where AoA is the estimated azimuth and elevation angles of arrival ( ), and AMP is the mean squared error of the estimated amplitude.

[0155] Figure 5 It shows that the statistical accuracy of the signal-to-noise ratio of this method in the range of [-10,20]dB is higher than that of the traditional SAGE algorithm and the low-computation maximum likelihood method.

[0156] (C) Computational complexity comparison:

[0157] For the SAGE algorithm, the main computational complexity is concentrated in the M step (maximization step), and the computational complexity is ,in is the dimension of the NMQ signal space (number of transmitting array elements × number of receiving array elements × number of frequency points), and the complexity of calculating the inner product is square, is the number of times the likelihood function is calculated for each parameter dimension, and K is the number of multipaths.

[0158] The computational complexity of MLE is ,in is the dimension of the NMQ signal space (number of transmitting array elements × number of receiving array elements × number of frequency points), and the complexity of calculating the inner product is square, is the number of times the likelihood function is calculated for each parameter dimension, i is the dimension of the parameter, and K is the number of multipaths.

[0159] For this method, the computational complexity is ,in is the computational complexity of training the proxy model, is the computational complexity of the proxy model prediction once, and is the number of times the proxy model predicts. For Gaussian process regression, the computational complexity of training and prediction is related to the number of samples, and K is the number of multipaths. For the specific algorithm running time, see Table 1.

[0160] Table 1. Running time of specific algorithms

[0161]

[0162] It can be seen that this method significantly reduces the running time of estimating the multi-dimensional parameters of the channel subpaths.

[0163] The parameter estimation device, electronic device, computer program product, and computer-readable storage medium provided in the embodiments of the present invention are introduced below. The parameter estimation device, electronic device, computer program product, and computer-readable storage medium described below can be referenced to each other with the parameter estimation method described above.

[0164] Please refer to Figure 7 , Figure 7 A structural block diagram of a parameter estimation device provided by an embodiment of the present invention, the device may include:

[0165] The likelihood function determination module 701 is used to determine the likelihood function corresponding to the parameter to be estimated in the multidimensional parameters according to the latest estimated value corresponding to the multidimensional parameters of the channel subpath when entering the current iteration; the multidimensional parameters include signal amplitude and the parameter to be estimated, and the parameter to be estimated includes signal delay, two-dimensional departure angle and two-dimensional arrival angle;

[0166] A response value determination module 702 is used to obtain a parameter value by sampling in a parameter space corresponding to the parameter to be estimated, and input the parameter value into a likelihood function corresponding to the parameter to be estimated to obtain a response value;

[0167] The proxy model processing module 703 is used to train the proxy model of the parameter to be estimated using the parameter value and the response value, and update the estimated value of the parameter to be estimated using the trained proxy model;

[0168] A signal amplitude updating module 704 is used to update the estimated value of the signal amplitude using the latest estimated value corresponding to the parameter to be estimated;

[0169] The iteration control module 705 is used to determine whether to end the iteration according to the preset iteration end condition, and enter the next round of iteration when it is determined not to exit the iteration, or output the latest corresponding estimated value of the multi-dimensional parameter when it is determined to exit the iteration.

[0170] Optionally, the proxy model processing module 703 may include:

[0171] The model building submodule is used to build and train the proxy model using parameter values ​​and response values ​​based on the Gaussian process regression method;

[0172] The model optimization submodule is used to search for the maximum value of the parameter to be estimated based on the gradient descent algorithm in the trained proxy model, and use the maximum value of the parameter to be estimated as the estimated value for updating the parameter to be estimated.

[0173] Optionally, the response value determination module 702 may include:

[0174] The sampling submodule is used to obtain parameter values ​​by sampling in the parameter space corresponding to the parameter to be estimated using the Latin hypercube method.

[0175] Optionally, the device may further include:

[0176] The initialization module is used to initialize the estimated values ​​of the multidimensional parameters when determining that the current iteration is the first iteration.

[0177] Optionally, the device may further include:

[0178] The estimation completion judgment module is used to judge whether the multidimensional parameters of each channel sub-path have completed the update of the estimated values ​​in this round of iteration; if so, enter the step of judging whether to end the iteration according to the preset iteration end condition; if not, for the next channel sub-path that has not completed the estimated value update, enter the step of determining the likelihood function corresponding to the parameter to be estimated in the multidimensional parameters based on the latest corresponding estimated value of the multidimensional parameters of the channel sub-path.

[0179] Optionally, the iteration control module 705 may include:

[0180] The iteration round judgment submodule is used to judge whether the executed iteration rounds have reached a preset upper limit value; if the executed iteration rounds have reached the preset upper limit value, it is determined to exit the iteration; if the executed iteration rounds have not reached the preset upper limit value, it is determined not to exit the iteration;

[0181] Or, a convergence judgment submodule is used to judge whether the difference between the estimated value of the parameter to be estimated determined in this round of iteration and its estimated value determined in the previous round of iteration is less than a preset threshold; if the difference is less than the preset threshold, it is determined to exit the iteration; if the difference is not less than the preset threshold, it is determined not to exit the iteration.

[0182] Optionally, the likelihood function determination module 701 may include:

[0183] An expected signal value determination submodule, used to determine the expected signal value corresponding to the channel subpath in this iteration according to the latest corresponding estimated value of the multi-dimensional parameter of the channel subpath;

[0184] The likelihood function determination submodule is used to determine the likelihood function corresponding to the parameter to be estimated by using the expected signal value and the latest estimated value corresponding to the parameter to be estimated.

[0185] Optionally, the likelihood function determination submodule may include:

[0186] A delay likelihood function determination unit, used to determine a likelihood function corresponding to a signal delay using an expected signal value, a latest estimate corresponding to a two-dimensional departure angle, and a latest estimate corresponding to a two-dimensional arrival angle;

[0187] A two-dimensional angle of departure likelihood function determination unit, configured to determine a likelihood function corresponding to the two-dimensional angle of departure using an expected signal value, a latest corresponding estimate of the signal delay, and a latest corresponding estimate of the two-dimensional angle of arrival when the estimated value of the signal delay is updated in this round of iteration;

[0188] The two-dimensional arrival angle likelihood function determination unit is used to determine the likelihood function corresponding to the two-dimensional arrival angle using the expected signal value, the latest corresponding estimate of the signal delay and the latest corresponding estimate of the two-dimensional departure angle when the estimated value of the two-dimensional departure angle is updated in this round of iteration.

[0189] Please refer to Figure 8 , Figure 8 This is a structural block diagram of an electronic device provided in an embodiment of the present invention. The embodiment of the present invention provides an electronic device 10, including a processor 11 and a memory 12; wherein the memory 12 is used to store a computer program; and the processor 11 is used to execute the parameter estimation method provided in the aforementioned embodiment when executing the computer program.

[0190] For the specific process of the above parameter estimation method, reference may be made to the corresponding contents provided in the aforementioned embodiments, which will not be elaborated here.

[0191] Furthermore, the memory 12 as a carrier for storing resources may be a read-only memory, a random access memory, a magnetic disk or an optical disk, etc., and the storage method may be temporary storage or permanent storage.

[0192] In addition, the electronic device 10 also includes a power supply 13, a communication interface 14, an input / output interface 15 and a communication bus 16; wherein the power supply 13 is used to provide working voltage for each hardware device on the electronic device 10; the communication interface 14 can create a data transmission channel between the electronic device 10 and an external device, and the communication protocol it follows is any communication protocol that can be applied to the technical solution of the present invention, and is not specifically limited here; the input / output interface 15 is used to obtain external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0193] An embodiment of the present invention further provides a computer program product, including a computer program / instruction, which implements the parameter estimation method described in the above embodiment when executed by a processor.

[0194] Since the embodiments of the computer program product part correspond to the embodiments of the parameter estimation method part, please refer to the description of the embodiments of the parameter estimation method part for the embodiments of the computer program product part, which will not be repeated here.

[0195] An embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the parameter estimation method described in the above embodiment is implemented.

[0196] Since the embodiments of the computer-readable storage medium part correspond to the embodiments of the parameter estimation method part, the embodiments of the storage medium part refer to the description of the embodiments of the parameter estimation method part, which will not be repeated here.

[0197] The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.

[0198] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0199] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0200] The parameter estimation method, device, electronic device, storage medium and program product provided by the present invention are introduced in detail above. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. It should be pointed out that for ordinary technicians in this technical field, without departing from the principles of the present invention, several improvements and modifications can be made to the present invention, and these improvements and modifications also fall within the scope of protection of the present invention.

Claims

1. A parameter estimation method, characterized in that: include: When entering this round of iteration, the likelihood function corresponding to the parameter to be estimated in the multidimensional parameters is determined according to the latest estimated value corresponding to the multidimensional parameters of the channel subpath; The multidimensional parameters include signal amplitude and the parameters to be estimated, and the parameters to be estimated include signal delay, two-dimensional departure angle and two-dimensional arrival angle; Sampling in the parameter space corresponding to the parameter to be estimated to obtain a parameter value, and inputting the parameter value into the likelihood function corresponding to the parameter to be estimated to obtain a response value; Training a proxy model of the parameter to be estimated using the parameter value and the response value, and updating an estimated value of the parameter to be estimated using the trained proxy model; Updating the estimated value of the signal amplitude using the latest estimated value corresponding to the parameter to be estimated; Whether to end the iteration is determined according to the preset iteration end condition, and when it is determined not to exit the iteration, the next round of iteration is entered, or when it is determined to exit the iteration, the latest corresponding estimated value of the multidimensional parameter is output.

2. The parameter estimation method according to claim 1, characterized in that: The step of training the proxy model of the parameter to be estimated by using the parameter value and the response value, and updating the estimated value of the parameter to be estimated by using the trained proxy model, comprises: Based on a Gaussian process regression method, the proxy model is established and trained using the parameter values ​​and the response values; In the trained proxy model, the maximum value of the parameter to be estimated is searched based on the gradient descent algorithm, and the maximum value of the parameter to be estimated is used as the estimated value for updating the parameter to be estimated.

3. The parameter estimation method according to claim 1, characterized in that: The sampling in the parameter space corresponding to the parameter to be estimated to obtain the parameter value includes: The parameter value is obtained by sampling in the parameter space corresponding to the parameter to be estimated using the Latin hypercube method.

4. The parameter estimation method according to claim 1, characterized in that: Before determining whether to end the iteration according to the preset iteration end condition, the following is also included: Determining whether the multidimensional parameters of each of the channel subpaths have completed the update of estimated values ​​in this round of iteration; If yes, then enter the step of determining whether to end the iteration according to the preset iteration end condition; If not, for the next channel sub-path for which the estimated value update has not been completed, the step of determining the likelihood function corresponding to the parameter to be estimated in the multi-dimensional parameters according to the latest estimated value corresponding to the multi-dimensional parameters of the channel sub-path is entered.

5. The parameter estimation method according to any one of claims 1 to 4, characterized in that: The determining, according to the latest estimated value corresponding to the multidimensional parameter of the channel subpath, a likelihood function corresponding to the parameter to be estimated in the multidimensional parameter comprises: Determining the expected signal value corresponding to the channel subpath in this iteration according to the latest corresponding estimated value of the multidimensional parameter of the channel subpath; The likelihood function corresponding to the parameter to be estimated is determined by using the expected signal value and the latest estimated value corresponding to the parameter to be estimated.

6. The parameter estimation method according to claim 5, characterized in that: The method of determining the likelihood function corresponding to the parameter to be estimated by using the expected signal value and the latest estimated value corresponding to the parameter to be estimated includes: Determine a likelihood function corresponding to the signal delay by using the expected signal value, the latest estimated value corresponding to the two-dimensional departure angle, and the latest estimated value corresponding to the two-dimensional arrival angle; When the estimated value of the signal delay is updated in this round of iteration, the likelihood function corresponding to the two-dimensional departure angle is determined using the expected signal value, the latest estimated value corresponding to the signal delay, and the latest estimated value corresponding to the two-dimensional arrival angle; When the estimated value of the two-dimensional departure angle is updated in this round of iteration, the likelihood function corresponding to the two-dimensional arrival angle is determined using the expected signal value, the latest estimated value corresponding to the signal delay and the latest estimated value corresponding to the two-dimensional departure angle.

7. A parameter estimation device, characterized in that: include: A likelihood function determination module, used to determine, when entering this round of iteration, the likelihood function corresponding to the parameter to be estimated in the multidimensional parameters according to the latest corresponding estimated value of the multidimensional parameters of the channel subpath; the multidimensional parameters include the signal amplitude and the parameter to be estimated, and the parameter to be estimated includes the signal delay, the two-dimensional departure angle and the two-dimensional arrival angle; A response value determination module, used for sampling in the parameter space corresponding to the parameter to be estimated to obtain a parameter value, and inputting the parameter value into the likelihood function corresponding to the parameter to be estimated to obtain a response value; A proxy model processing module, used to train the proxy model of the parameter to be estimated using the parameter value and the response value, and update the estimated value of the parameter to be estimated using the trained proxy model; A signal amplitude updating module, used to update the estimated value of the signal amplitude using the latest estimated value corresponding to the parameter to be estimated; The iteration control module is used to determine whether to end the iteration according to the preset iteration end condition, and enter the next round of iteration when it is determined not to exit the iteration, or output the latest corresponding estimated value of the multidimensional parameter when it is determined to exit the iteration.

8. An electronic device, characterized in that: include: Memory for storing computer programs; A processor, configured to implement the parameter estimation method according to any one of claims 1 to 6 when executing the computer program.

9. A computer program product comprising a computer program or instructions, characterized in that When the computer program or instruction is executed by a processor, the parameter estimation method according to any one of claims 1 to 6 is implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are loaded and executed by a processor, the parameter estimation method according to any one of claims 1 to 6 is implemented.

Images