A method for determining a time delay and related products

Abstract

The application discloses a time delay determination method and related products. In the scheme, based on the to-be-processed signal data, an initial dictionary matrix is constructed, and based on the to-be-processed signal data and the initial dictionary matrix, an initial time delay result corresponding to the to-be-processed signal is calculated. Based on the initial time delay result, the initial dictionary matrix is adjusted to obtain a target dictionary matrix. Based on the to-be-processed signal data and the target dictionary matrix, a target time delay result is calculated. Compared with the low accuracy of the existing time delay determination method, the application has obvious advantages.

Description

Technical Field

[0001] This application relates to the field of signal processing, and in particular to a method for determining time delay and related products. Background Technology

[0002] In modern signal processing systems such as wireless communication, radar detection, and sonar positioning, latency determination is a core fundamental technology, and the accuracy of the determined latency directly affects system performance. With the development of technologies such as 5G / 6G communication, autonomous driving, and intelligent sensing, the requirements for latency accuracy are becoming increasingly stringent. Latency refers to the time it takes for a signal to travel from a transmitter, such as a base station, tag, or radar, to a receiver. By determining the latency, the distance between the transmitter and receiver can be deduced, thus enabling positioning.

[0003] Traditional methods for determining time delay are mainly divided into model-based methods and data-driven methods. Model-based methods rely on accurate physical channel models, which have clear physical meaning and good interpretability. However, as the size of antenna arrays increases and the channel environment becomes more complex, the computational complexity grows exponentially, making it difficult to meet real-time processing requirements and prone to latency issues, thus reducing the accuracy of the determined time delay. Data-driven methods learn the mapping relationship between input and output through a large amount of training data. Although this avoids complex matrix operations, the model has poor generalization ability, and changes in the scenario severely affect the accuracy of the determined time delay.

[0004] In conclusion, improving the accuracy of a given time delay is a technical problem that urgently needs to be solved. Summary of the Invention

[0005] To address the aforementioned issues, this application provides a method for determining time delay and related products, with the aim of improving the accuracy of the determined time delay and thereby achieving accurate positioning.

[0006] The embodiments of this application disclose the following technical solutions:

[0007] The first aspect of this application provides a method for determining time delay, the method comprising:

[0008] Based on the signal data to be processed, an initial dictionary matrix is ​​constructed, and based on the signal data to be processed and the initial dictionary matrix, the initial time delay result corresponding to the signal to be processed is calculated.

[0009] Based on the initial delay result, the initial dictionary matrix is ​​adjusted to obtain the target dictionary matrix;

[0010] Based on the signal data to be processed and the target dictionary matrix, the target time delay result is calculated.

[0011] Optionally, constructing the initial dictionary matrix based on the signal data to be processed includes:

[0012] Based on prior channel information, the power delay distribution of the time delay is determined;

[0013] Based on the power delay distribution of the delay, the range of values ​​for the delay is determined;

[0014] The value range is divided into multiple sub-ranges, the cumulative energy value of each sub-range is calculated, and the multiple sub-ranges are classified based on the cumulative energy value of each sub-range.

[0015] If the first sub-interval is of the first category, then sampling points are set within the first sub-interval according to the first density; if the first sub-interval is of the second category, then sampling points are set within the first sub-interval according to the second density; the first sub-interval is any sub-interval among the plurality of sub-regions; the first density is higher than the second density;

[0016] Based on the sampling point settings of the multiple sub-regions, an initial dictionary matrix is ​​constructed.

[0017] Optionally, adjusting the initial dictionary matrix based on the initial delay result to obtain the target dictionary matrix includes:

[0018] Calculate the residual energy value corresponding to the initial time delay result;

[0019] If the residual energy value exceeds a first preset threshold, the preset range of the time delay is divided into multiple sub-intervals, the residual energy value of each sub-interval is calculated, and the initial dictionary matrix is ​​adjusted based on the residual energy value of each sub-interval to obtain the target dictionary matrix; otherwise, the initial dictionary matrix is ​​used as the target dictionary matrix.

[0020] Optionally, adjusting the initial dictionary matrix based on the residual energy value of each sub-interval to obtain the target dictionary matrix includes:

[0021] Determine whether the residual energy value of each sub-interval exceeds a second preset threshold to obtain a first determination result;

[0022] If the first judgment result is yes, then the corresponding sub-interval is marked as the sub-interval to be encrypted, and the sampling density of the sub-interval to be encrypted is adjusted to obtain the adjusted sub-interval; based on the adjusted sub-interval, the initial dictionary matrix is ​​adjusted to obtain the target dictionary matrix;

[0023] If the first judgment result is negative, the corresponding sub-interval will not be adjusted.

[0024] Optionally, adjusting the initial dictionary matrix based on the adjusted sub-intervals to obtain the target dictionary matrix includes:

[0025] Based on the adjusted sub-intervals, the initial dictionary matrix is ​​adjusted to obtain the adjusted dictionary matrix;

[0026] Set the value of k to 1; use the adjusted dictionary matrix as the initial data for the k-th iteration;

[0027] Based on the signal data to be processed and the initial data of the k-th iteration, the time delay value of the k-th iteration is calculated;

[0028] Determine whether the time delay value of the k-th iteration satisfies the first iteration termination condition, and obtain the second determination result;

[0029] If the second judgment result is negative, the initial data of the kth iteration is adjusted to obtain the initial data of the (k+1)th iteration, the value of k is incremented by 1, and the step of calculating the time delay value of the kth iteration based on the signal data to be processed and the initial data of the kth iteration is returned.

[0030] If the second judgment result is yes, then the iteration ends, and the initial data of the kth iteration is output as the target dictionary matrix.

[0031] Optionally, calculating the target time delay result based on the signal data to be processed and the target dictionary matrix includes:

[0032] The signal data to be processed and the target dictionary matrix are processed to obtain the posterior distribution of the time delay; the posterior distribution of the time delay includes a first hyperparameter and a second hyperparameter;

[0033] The first hyperparameter and the second hyperparameter are iteratively optimized by a deep unfolded network to obtain the target mean and target covariance of the time delay.

[0034] Based on the target mean and the target covariance, the posterior distribution of the time delay is updated, and the target time delay result is determined based on the updated posterior distribution of the time delay.

[0035] Optionally, the iterative optimization of the first and second hyperparameters using a deep unfolded network to obtain the target mean and target covariance of the time delay includes:

[0036] Set the initial values ​​for the first hyperparameter and the second hyperparameter;

[0037] Set the value of k to 1; use the initial values ​​of the first hyperparameter and the second hyperparameter as the initial data for the kth iteration;

[0038] Based on the initial data of the k-th iteration, the mean and covariance of the k-th iteration are calculated.

[0039] Based on the mean and covariance of the k-th iteration, the initial data of the k-th iteration is updated to obtain the initial data of the (k+1)-th iteration.

[0040] Determine whether the initial data of the (k+1)th iteration satisfies the second iteration termination condition to obtain the third determination result;

[0041] If the third judgment result is negative, then the initial data of the (k+1)th iteration is adjusted to obtain the initial data of the (k+2)th iteration, the value of k is incremented by 1, and the step of calculating the mean and covariance of the kth iteration based on the initial data of the kth iteration is returned.

[0042] If the third judgment result is yes, the iteration ends, and the target mean and target covariance of the time delay are calculated based on the initial data of the k-th iteration.

[0043] Optionally, the step of processing the signal data to be processed and the target dictionary matrix to obtain the posterior distribution of the time delay includes:

[0044] Based on the signal data to be processed and the target dictionary matrix, the likelihood probability distribution is calculated;

[0045] Based on the automatic correlation decision method, the prior probability distribution of time delay is determined;

[0046] Based on the likelihood probability distribution and the prior probability distribution, the marginal likelihood probability distribution of the time delay is calculated;

[0047] Based on the likelihood probability distribution, the prior probability distribution, and the marginal likelihood probability distribution, the posterior distribution of the time delay is calculated.

[0048] A second aspect of this application provides a time delay determination apparatus, the apparatus comprising:

[0049] The initial delay result determination module is used to construct an initial dictionary matrix based on the signal data to be processed, and to calculate the initial delay result corresponding to the signal to be processed based on the signal data to be processed and the initial dictionary matrix.

[0050] The dictionary matrix adjustment module is used to adjust the initial dictionary matrix based on the initial delay result to obtain the target dictionary matrix;

[0051] The target delay result determination module is used to calculate the target delay result based on the signal data to be processed and the target dictionary matrix.

[0052] A third aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements a delay determination method as provided in any implementation of the first aspect.

[0053] Compared with the prior art, this application has the following advantages:

[0054] The delay determination method provided in this application constructs an initial dictionary matrix based on the signal data to be processed, and calculates the initial delay result corresponding to the signal based on the signal data to be processed and the initial dictionary matrix. Based on the initial delay result, the initial dictionary matrix is ​​adjusted to obtain a target dictionary matrix. Based on the signal data to be processed and the target dictionary matrix, the target delay result is calculated. By using the initial experimental results to guide the adjustment of the dictionary matrix to obtain the target dictionary matrix, the coherence between the target dictionary matrix and the actual signal is enhanced, enabling more accurate capture of the characteristics of the signal data to be processed and improving the accuracy of the delay result. Attached Figure Description

[0055] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0056] Figure 1 A flowchart illustrating a method for determining latency provided in an embodiment of this application;

[0057] Figure 2 This is a schematic diagram of a delay determination device provided in an embodiment of this application. Detailed Implementation

[0058] As described earlier, traditional methods for determining time delay are mainly divided into model-based methods and data-driven methods. Model-based methods rely on accurate physical channel models and achieve parameter estimation through mathematical optimization techniques such as maximum likelihood estimation or subspace decomposition. This method has clear physical meaning and good interpretability. However, as the size of the antenna array increases and the channel environment becomes more complex, its computational complexity grows exponentially, making it difficult to meet real-time processing requirements and prone to problems such as positioning delay, thus reducing the accuracy of the determined time delay. Data-driven methods learn the mapping relationship between input and output through a large amount of training data. Although this avoids complex matrix operations, the model has poor generalization ability, and changes in the scene seriously affect the accuracy of the determined time delay.

[0059] Existing methods fail to effectively combine channel prior information with data-driven approaches, resulting in a closed-loop optimization between dictionary matrix construction and Bayesian inference.

[0060] In view of the above problems, this application proposes a method for determining time delay and related products. Based on the signal data to be processed, an initial dictionary matrix is ​​constructed, and based on the signal data to be processed and the initial dictionary matrix, an initial time delay result corresponding to the signal to be processed is calculated; based on the initial time delay result, the initial dictionary matrix is ​​adjusted to obtain a target dictionary matrix; based on the signal data to be processed and the target dictionary matrix, a target time delay result is calculated.

[0061] The delay determination method provided in this application constructs an initial dictionary matrix based on the signal data to be processed, and calculates the initial delay result corresponding to the signal based on the signal data to be processed and the initial dictionary matrix. Based on the initial delay result, the initial dictionary matrix is ​​adjusted to obtain a target dictionary matrix. Based on the signal data to be processed and the target dictionary matrix, the target delay result is calculated. By using the initial experimental results to guide the adjustment of the dictionary matrix to obtain the target dictionary matrix, the coherence between the target dictionary matrix and the actual signal is enhanced, which can more accurately capture the characteristics of the signal data to be processed, improve the accuracy of the delay result, and reduce the computational complexity.

[0062] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0063] See Figure 1 The figure is a flowchart of a delay determination method provided in an embodiment of this application. Figure 1 As shown, the method includes the following steps:

[0064] S101. Based on the signal data to be processed, construct an initial dictionary matrix, and based on the signal data to be processed and the initial dictionary matrix, calculate the initial time delay result corresponding to the signal to be processed.

[0065] The initial time delay result here is a point estimate.

[0066] In one feasible implementation:

[0067] Based on prior channel information, the power delay distribution of the time delay is obtained.

[0068] Based on the power delay distribution of the delay, the range of values ​​for the delay is determined.

[0069] The value range is divided into multiple sub-ranges, the cumulative energy value of each sub-range is calculated, and the multiple sub-ranges are classified based on the cumulative energy value of each sub-range.

[0070] If the first sub-interval is of the first category, then sampling points are set within the first sub-interval according to the first density; if the first sub-interval is of the second category, then sampling points are set within the first sub-interval according to the second density; the first sub-interval is any one of the plurality of sub-intervals; the first density is higher than the second density.

[0071] Based on the sampling point settings of the multiple sub-regions, an initial dictionary matrix is ​​constructed.

[0072] The prior channel information includes indoor environmental geometry, material reflection characteristics, and historical channel measurement data. The prior channel information is updated based on historical data.

[0073] The method for determining the power delay distribution is not restricted here; for example, it can be determined through a constructed power delay distribution model. Based on the power delay distribution, the delay interval containing effective multipath is taken as the range of delay values, for example, the interval corresponding to 99% energy coverage.

[0074] There are no restrictions on how the value interval is divided into multiple sub-intervals, such as equal intervals.

[0075] The first category and the second category are divided based on a preset cumulative energy threshold. The energy corresponding to the first category is greater than or equal to the preset cumulative energy threshold, and it is a high-energy category. The probability value corresponding to the second category is less than the preset cumulative energy threshold, and it is a low-energy category.

[0076] Both the first density and the second density are preset, with the first density being higher than the second density. For example, setting the sampling points according to the first density allows for one sampling point to be set every 0.1 ns, while setting the sampling points according to the second density allows for one sampling point to be set every 5 ns.

[0077] By introducing prior channel information to determine the power delay distribution, and dividing and classifying the delay value range according to the power delay distribution, different sub-intervals correspond to different sampling densities. This non-uniform grid design driven by prior channel information uses a dense grid in the high-energy range and a sparse grid in the weak-energy range, achieving the matching between the dictionary atom distribution and the real multipath energy distribution, and significantly reducing the redundancy of the initial dictionary matrix.

[0078] S102. Based on the initial delay result, adjust the initial dictionary matrix to obtain the target dictionary matrix.

[0079] In one feasible implementation:

[0080] Calculate the residual energy value corresponding to the initial time delay result.

[0081] If the residual energy value exceeds a first preset threshold, the preset range of the time delay is divided into multiple sub-intervals, the residual energy value of each sub-interval is calculated, and the initial dictionary matrix is ​​adjusted based on the residual energy value of each sub-interval to obtain the target dictionary matrix; otherwise, the initial dictionary matrix is ​​used as the target dictionary matrix.

[0082] The first preset threshold is used to determine whether to start the dictionary matrix optimization process, and can be determined based on the noise power of the received signal. The larger the residual energy value, the more likely the initial dictionary matrix has failed to capture the multipath components in the signal, and the initial dictionary matrix needs to be adjusted; the smaller the residual energy, the more likely the residual is mainly caused by noise or modeling error, and the initial dictionary matrix does not need to be adjusted.

[0083] The preset range of the delay value is determined based on the maximum transmission delay of the signal, which is determined based on the transmission power of the space or device.

[0084] By calculating the residual energy of the initial time delay result, the modeling error of the initial dictionary matrix is ​​quantified, and the potential multipath region that has not been modeled is accurately located, thus realizing the adaptive optimization of the dictionary matrix.

[0085] Optionally, based on the residual energy value of each sub-interval, the initial dictionary matrix is ​​adjusted to obtain the target dictionary matrix, including:

[0086] Determine whether the residual energy value of each sub-interval exceeds a second preset threshold to obtain a first determination result.

[0087] If the first judgment result is yes, the corresponding sub-interval is marked as the sub-interval to be encrypted, and the sampling density of the sub-interval to be encrypted is adjusted to obtain the adjusted sub-interval; based on the adjusted sub-interval, the initial dictionary matrix is ​​adjusted to obtain the target dictionary matrix. If the first judgment result is no, the corresponding sub-interval is not adjusted.

[0088] The second preset threshold is used to determine which areas need to be encrypted at the sub-interval level. The second preset threshold is less than the first preset threshold.

[0089] Optionally, adjusting the sampling density of the sub-interval to be encrypted includes adjusting the sampling interval of the sub-interval to be encrypted to half of the original.

[0090] Optionally, based on the adjusted sub-intervals, the initial dictionary matrix is ​​adjusted to obtain the target dictionary matrix, including:

[0091] Based on the adjusted sub-intervals, the initial dictionary matrix is ​​adjusted to obtain the adjusted dictionary matrix.

[0092] Set the value of k to 1; use the adjusted dictionary matrix as the initial data for the k-th iteration.

[0093] Based on the signal data to be processed and the initial data of the k-th iteration, the time delay result of the k-th iteration is calculated.

[0094] Determine whether the delay result of the k-th iteration satisfies the first iteration termination condition to obtain the second determination result.

[0095] If the second judgment result is negative, the initial data of the kth iteration is adjusted to obtain the initial data of the (k+1)th iteration, the value of k is incremented by 1, and the step of calculating the time delay result of the kth iteration based on the signal data to be processed and the initial data of the kth iteration is returned.

[0096] If the second judgment result is yes, then the iteration ends, and the initial data of the kth iteration is output as the target dictionary matrix.

[0097] The first iteration ends when the maximum number of iterations is reached or the residual energy value corresponding to the delay result is lower than the first preset threshold.

[0098] By judging the residual energy value of each sub-interval, the sub-interval to be encrypted is accurately identified. Only the sampling density of the sub-interval to be encrypted is adjusted, realizing on-demand encryption and local optimization. While ensuring the accuracy of the determined delay result, the computational complexity of the dictionary matrix is ​​reduced.

[0099] S103. Based on the signal data to be processed and the target dictionary matrix, the target time delay result is calculated.

[0100] In one feasible implementation:

[0101] The signal data to be processed and the target dictionary matrix are processed to obtain the posterior distribution of the time delay; the posterior distribution of the time delay includes a first hyperparameter and a second hyperparameter.

[0102] The first and second hyperparameters are iteratively optimized by using a deep unfolded network to obtain the target mean and target covariance of the time delay.

[0103] Based on the target mean and the target covariance, the posterior distribution of the time delay is updated, and the target time delay result is determined based on the updated posterior distribution of the time delay.

[0104] The target delay result here is a point estimate.

[0105] Deep unfolded networks expand the traditional Bayesian iterative algorithm into a neural network structure, combining the advantages of model-driven and data-driven approaches. Each layer corresponds to one iterative update, and each layer has a clear physical meaning, such as updating the mean or updating the precision, making it highly interpretable. Moreover, deep unfolded networks can achieve adaptive updates through learning, significantly reducing the number of iterations and improving the accuracy of deterministic time delay results.

[0106] Optionally, the first hyperparameter and the second hyperparameter are iteratively optimized using a deep unfolded network to obtain the target mean and target covariance of the time delay, including:

[0107] Set the initial values ​​for the first hyperparameter and the second hyperparameter.

[0108] Set the value of k to 1; use the initial values ​​of the first hyperparameter and the second hyperparameter as the initial data for the kth iteration.

[0109] Based on the initial data of the k-th iteration, the mean and covariance of the k-th iteration are calculated.

[0110] Based on the mean and covariance of the k-th iteration, the initial data of the k-th iteration is updated to obtain the initial data of the (k+1)-th iteration.

[0111] Determine whether the initial data of the (k+1)th iteration satisfies the second iteration termination condition to obtain the third determination result.

[0112] If the third judgment result is negative, then the initial data of the (k+1)th iteration is adjusted to obtain the initial data of the (k+2)th iteration, the value of k is incremented by 1, and the step of calculating the mean and covariance of the kth iteration based on the initial data of the kth iteration is returned.

[0113] If the third judgment result is yes, the iteration ends, and the target mean and target covariance of the time delay are calculated based on the initial data of the k-th iteration.

[0114] The second iteration ends when the change in the mean and the change in the covariance of the k-th iteration are both less than a preset threshold.

[0115] For example, the above iterative optimization process can be achieved through a deep unfolded network:

[0116] definition For the entire learning process, l represents the number of network layers, and y represents the input signal. This represents the parameters that need to be learned in each iteration, in order to determine the specific parameters for each layer. The learning process corresponds to the iterative process of updating hyperparameters using the EM method, which divides the iteration into two stages: f l and g l .

[0117] f l Process: f l It receives the mean and covariance from layer l-1 and updates the hyperparameters using the following formula:

[0118] ;

[0119] ;

[0120] Where, α i These are the hyperparameter values ​​for this layer, controlling the i-th parameter x. i Prior distribution intensity; α i-1 It is the hyperparameter value of the previous layer, w i It is the i-th component of the posterior mean; It is the i-th diagonal element of the posterior covariance matrix; σ 2 It is the noise variance, which describes the level of random noise in the signal data y to be processed; This is the actual observation error; y is the prediction variance; y is the vector corresponding to the signal data to be processed; A is the dictionary matrix; x is the time delay to be determined; N is the number of data points.

[0121] g l Process: g l Is receiving from f l of The mean and covariance of this layer are calculated using the following formula:

[0122] ;

[0123] ;

[0124] in, Σx is the posterior mean, used to determine the optimal value of the time delay x; Σx is the posterior covariance, describing the uncertainty of the time delay; A is the dictionary matrix. Reflects the structure of the dictionary matrix; σ 2 It is the noise variance; diag(α) is a diagonal matrix composed of α, representing the inverse of the prior covariance.

[0125] f lThe process is implemented using a deep network. The specific steps are as follows: (1) The input y first passes through the first convolutional layer Conv1, applies the ReLU activation function, and performs max pooling. (2) Then it passes through the second convolutional layer Conv2, which also applies the ReLU activation function. The first and second convolutional layers are connected by a residual connection to solve the degradation problem in deep network training. (3) The three-dimensional feature map tensor is flattened into a one-dimensional tensor using Flatten. (4) Finally, it passes through the first fully connected layer FC1, applies the ReLU activation function, and then passes through the second fully connected layer FC2 (output layer) to obtain the final result.

[0126] g l The process is implemented using a traditional iterative method. It receives the output from the previous layer.

[0127] The current mean and covariance are calculated, and the results are used as input values ​​for the next layer of the deep network. Traditional iterative methods include singular value decomposition, principal component analysis, or nonnegative matrix factorization.

[0128] Here, cross-entropy loss is used to calculate the loss function during the iterative optimization process.

[0129] Optionally, the method further includes:

[0130] The output of the deep unfolded network is filtered.

[0131] In one feasible implementation, an SG filter (Savitzky-Golay Filter) is added after the deep unfolded network to smooth the output. The SG filter fits a continuous subset of adjacent data points to a low-order polynomial using linear least squares; exemplarily, the application window length is 71 and the polynomial order is 5. The SG filter is a digital filter that can be applied to a set of data to smooth it, improving data accuracy without altering signal trends or width.

[0132] Optionally, the signal data to be processed and the target dictionary matrix are processed to obtain the posterior distribution of the time delay, including:

[0133] Based on the signal data to be processed and the target dictionary matrix, the likelihood probability distribution is calculated.

[0134] Based on the automatic correlation decision method, the prior probability distribution of time delay is determined.

[0135] Based on the likelihood probability distribution and the prior probability distribution, the marginal likelihood probability distribution of the time delay is calculated.

[0136] Based on the likelihood probability distribution, the prior probability distribution, and the marginal likelihood probability distribution, the posterior distribution of the time delay is calculated.

[0137] For example, the likelihood probability distribution The calculation formula is as follows:

[0138] ;

[0139] Where y is the vector corresponding to the signal data to be processed; A is the dictionary matrix; and x is the time delay to be determined. It is a normalization constant; It is an exponential function; It is the sum of squared residuals, used to quantify the degree of difference between actual and predicted values; σ 2 is the noise variance; N is the number of data points.

[0140] In sparse Bayesian learning, in order to give each parameter x i Independent processing is performed using Automatic Relevance Determination (ARD) priors. ARD achieves sparsity to enhance time delay resolution. The ARD prior introduces a hyperparameter vector α = (α1, α2, α3…α…). M ), for each x i Configure an independent variance. When α i As x approaches infinity, i The sparsity is achieved by decomposing the complex matrix into a real matrix. Since Channel State Information (CSI) data is mostly complex, its computational complexity is high, and complex domain signals are difficult to process in deep learning frameworks. Therefore, all complex-valued matrices are decomposed into real-valued matrices. The time delay to be determined, x, is set to a Gaussian distribution with a mean of 0, and for each x... i Configure an independent variance to obtain the prior probability distribution of x. :

[0141] ;

[0142] Where each component x i Independently obeys N(0, This means that each parameter has an independent variance. ; Indicated by A diagonal matrix with the main diagonal.

[0143] Marginal likelihood probability distribution through The calculation involves integrating the entire x over the parameter space, meaning taking into account every possible value of x, to automatically balance fitting accuracy and complexity. Substituting the expressions for the prior distribution and the likelihood function into the calculation of the likelihood probability distribution yields the likelihood probability distribution. :

[0144] ;

[0145] Where y is the vector corresponding to the signal data to be processed; A is the dictionary matrix; and x is the time delay to be determined. It is a normalization constant; It is an exponential function; It is the sum of squared residuals, used to quantify the degree of difference between actual and predicted values; σ 2 This represents the noise variance; N is the number of data points. It is a diagonal matrix, representing the inverse of the prior covariance; It is an inverse diagonal matrix, representing the prior covariance matrix; This represents the prior penalty term; the larger the value, the more complex the term.

[0146] Likelihood probability distribution The exponential parts are merged, and then the terms in the exponent are expanded into a quadratic form in x, finally yielding the marginal likelihood probability distribution. for:

[0147] ;

[0148] in, , Represents the effective covariance matrix. It is observation noise. This reflects prior uncertainty; This represents the effective precision matrix.

[0149] Based on Bayes' theorem, the posterior distribution is calculated using the following formula. :

[0150] ;

[0151] in, Represents the likelihood probability distribution. Let x represent the prior probability distribution. This represents the marginal likelihood probability distribution.

[0152] Substituting the prior probability distribution, likelihood probability distribution, and marginal likelihood probability distribution into the equation, we can calculate the posterior distribution:

[0153] ;

[0154] in, The amount of information from the signal data to be processed reflects the degree of influence of the observed value on x; Derived from prior knowledge, reflecting the confidence level in x; Indicates the posterior covariance; This represents the posterior mean.

[0155] The delay determination method provided in this application involves constructing an initial dictionary matrix based on the signal data to be processed, and calculating the initial delay result corresponding to the signal based on the signal data to be processed and the initial dictionary matrix. Based on the initial delay result, the initial dictionary matrix is ​​adjusted to obtain a target dictionary matrix. Based on the signal data to be processed and the target dictionary matrix, the target delay result is calculated. By using the initial experimental results to guide the adjustment of the dictionary matrix to obtain the target dictionary matrix, the coherence between the target dictionary matrix and the actual signal is enhanced, enabling more accurate capture of the characteristics of the signal data to be processed and improving the accuracy of the delay result.

[0156] Compared with existing technologies, the dictionary matrix update strategy in this application is local area encryption. While improving the dictionary resolution, the computational burden is significantly reduced due to the reduced coverage area. It is suitable for indoor multipath channel arrival time estimation, and the multipath energy attenuation law is clear.

[0157] Based on the delay determination method described in the preceding embodiments, this application also provides a delay determination device. Figure 2 This is a schematic diagram of the device. Figure 2 As shown, the device for determining the time delay includes:

[0158] The initial delay result determination module 201 is used to construct an initial dictionary matrix based on the signal data to be processed, and to calculate the initial delay result corresponding to the signal to be processed based on the signal data to be processed and the initial dictionary matrix.

[0159] The dictionary matrix adjustment module 202 is used to adjust the initial dictionary matrix based on the initial delay result to obtain the target dictionary matrix.

[0160] The target delay result determination module 203 is used to calculate the target delay result based on the signal data to be processed and the target dictionary matrix.

[0161] Optionally, the initial delay result determination module is used to:

[0162] Based on prior channel information, the power delay distribution of the time delay is determined.

[0163] Based on the power delay distribution of the delay, the range of values ​​for the delay is determined.

[0164] The value range is divided into multiple sub-ranges, the cumulative energy value of each sub-range is calculated, and the multiple sub-ranges are classified based on the cumulative energy value of each sub-range.

[0165] If the first sub-interval is of the first category, then sampling points are set within the first sub-interval according to the first density; if the first sub-interval is of the second category, then sampling points are set within the first sub-interval according to the second density; the first sub-interval is any sub-interval among the plurality of sub-regions; the first density is higher than the second density.

[0166] Based on the sampling point settings of the multiple sub-regions, an initial dictionary matrix is ​​constructed.

[0167] Optionally, the dictionary matrix adjustment module is used for:

[0168] Calculate the residual energy value corresponding to the initial time delay result.

[0169] If the residual energy value exceeds a first preset threshold, the preset range of the time delay is divided into multiple sub-intervals, the residual energy value of each sub-interval is calculated, and the initial dictionary matrix is ​​adjusted based on the residual energy value of each sub-interval to obtain the target dictionary matrix; otherwise, the initial dictionary matrix is ​​used as the target dictionary matrix.

[0170] Optionally, adjusting the initial dictionary matrix based on the residual energy value of each sub-interval to obtain the target dictionary matrix includes:

[0171] Determine whether the residual energy value of each sub-interval exceeds a second preset threshold to obtain a first determination result.

[0172] If the first judgment result is yes, then the corresponding sub-interval is marked as the sub-interval to be encrypted, and the sampling density of the sub-interval to be encrypted is adjusted to obtain the adjusted sub-interval; based on the adjusted sub-interval, the initial dictionary matrix is ​​adjusted to obtain the target dictionary matrix.

[0173] If the first judgment result is negative, the corresponding sub-interval will not be adjusted.

[0174] Optionally, adjusting the initial dictionary matrix based on the adjusted sub-intervals to obtain the target dictionary matrix includes:

[0175] Based on the adjusted sub-intervals, the initial dictionary matrix is ​​adjusted to obtain the adjusted dictionary matrix.

[0176] Set the value of k to 1; use the adjusted dictionary matrix as the initial data for the k-th iteration.

[0177] Based on the signal data to be processed and the initial data of the k-th iteration, the time delay value of the k-th iteration is calculated.

[0178] Determine whether the time delay value of the k-th iteration satisfies the first iteration termination condition to obtain the second determination result.

[0179] If the second judgment result is negative, then the initial data of the kth iteration is adjusted to obtain the initial data of the (k+1)th iteration, the value of k is incremented by 1, and the step of calculating the time delay value of the kth iteration based on the signal data to be processed and the initial data of the kth iteration is returned.

[0180] If the second judgment result is yes, then the iteration ends, and the initial data of the kth iteration is output as the target dictionary matrix.

[0181] Optionally, the target delay result determination module is used for:

[0182] The signal data to be processed and the target dictionary matrix are processed to obtain the posterior distribution of the parameters to be processed; the posterior distribution of the time delay includes a first hyperparameter and a second hyperparameter.

[0183] The first and second hyperparameters are iteratively optimized by using a deep unfolded network to obtain the target mean and target covariance of the time delay.

[0184] Based on the target mean and the target covariance, the posterior distribution of the time delay is updated, and the target time delay result is determined based on the updated posterior distribution of the time delay.

[0185] Optionally, the iterative optimization of the first and second hyperparameters using a deep unfolded network to obtain the target mean and target covariance of the time delay includes:

[0186] Set the initial values ​​for the first hyperparameter and the second hyperparameter.

[0187] Set the value of k to 1; use the initial values ​​of the first hyperparameter and the second hyperparameter as the initial data for the kth iteration.

[0188] Based on the initial data of the k-th iteration, the mean and covariance of the k-th iteration are calculated.

[0189] Based on the mean and covariance of the k-th iteration, the initial data of the k-th iteration is updated to obtain the initial data of the (k+1)-th iteration.

[0190] Determine whether the initial data of the (k+1)th iteration satisfies the second iteration termination condition to obtain the third determination result.

[0191] If the third judgment result is negative, then the initial data of the (k+1)th iteration is adjusted to obtain the initial data of the (k+2)th iteration, the value of k is incremented by 1, and the step of calculating the mean and covariance of the kth iteration based on the initial data of the kth iteration is returned.

[0192] If the third judgment result is yes, the iteration ends, and the target mean and target covariance of the time delay are calculated based on the initial data of the k-th iteration.

[0193] Optionally, the step of processing the signal data to be processed and the target dictionary matrix to obtain the posterior distribution of the time delay includes:

[0194] Based on the signal data to be processed and the target dictionary matrix, the likelihood probability distribution is calculated.

[0195] Based on the automatic correlation decision method, the prior probability distribution of time delay is determined.

[0196] Based on the likelihood probability distribution and the prior probability distribution, the marginal likelihood probability distribution of the time delay is calculated.

[0197] Based on the likelihood probability distribution, the prior probability distribution, and the marginal likelihood probability distribution, the posterior distribution of the time delay is calculated.

[0198] Furthermore, embodiments of this application also provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements a delay determination method as described in any of the method embodiments.

[0199] It should be noted that the various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for the device embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method embodiments. The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separate. The components indicated 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 solution according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0200] The above description is merely one specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for determining time delay, characterized in that, The method includes: Based on the signal data to be processed, an initial dictionary matrix is ​​constructed, and based on the signal data to be processed and the initial dictionary matrix, the initial time delay result corresponding to the signal to be processed is calculated. Based on the initial delay result, the initial dictionary matrix is ​​adjusted to obtain the target dictionary matrix; Based on the signal data to be processed and the target dictionary matrix, the target time delay result is calculated; The step of adjusting the initial dictionary matrix based on the initial delay result to obtain the target dictionary matrix includes: Calculate the residual energy value corresponding to the initial time delay result; Determine whether the residual energy value exceeds a first preset threshold. If so, divide the preset range of the time delay into multiple sub-intervals, calculate the residual energy value of each sub-interval, and adjust the initial dictionary matrix based on the residual energy value of each sub-interval to obtain the target dictionary matrix. If not, use the initial dictionary matrix as the target dictionary matrix. The calculation of the target time delay result based on the signal data to be processed and the target dictionary matrix includes: The signal data to be processed and the target dictionary matrix are processed to obtain the posterior distribution of the time delay; the posterior distribution of the time delay includes a first hyperparameter and a second hyperparameter; The first hyperparameter and the second hyperparameter are iteratively optimized by a deep unfolded network to obtain the target mean and target covariance of the time delay. Based on the target mean and the target covariance, the posterior distribution of the time delay is updated, and the target time delay result is determined based on the updated posterior distribution of the time delay.

2. The method according to claim 1, characterized in that, The initial dictionary matrix is ​​constructed based on the signal data to be processed, including: Based on prior channel information, the power delay distribution of the time delay is determined; Based on the power delay distribution of the delay, the range of values ​​for the delay is determined; The value range is divided into multiple sub-ranges, the cumulative energy value of each sub-range is calculated, and the multiple sub-ranges are classified based on the cumulative energy value of each sub-range. If the first sub-interval is of the first category, then sampling points are set within the first sub-interval according to the first density; if the first sub-interval is of the second category, then sampling points are set within the first sub-interval according to the second density; the first sub-interval is any one of the plurality of sub-intervals; the first density is higher than the second density; Based on the sampling point settings of the multiple sub-intervals, an initial dictionary matrix is ​​constructed.

3. The method according to claim 1, characterized in that, The process of adjusting the initial dictionary matrix based on the residual energy value of each sub-interval to obtain the target dictionary matrix includes: Determine whether the residual energy value of each sub-interval exceeds a second preset threshold to obtain a first determination result; If the first judgment result is yes, then the corresponding sub-interval is marked as the sub-interval to be encrypted, and the sampling density of the sub-interval to be encrypted is adjusted to obtain the adjusted sub-interval; based on the adjusted sub-interval, the initial dictionary matrix is ​​adjusted to obtain the target dictionary matrix; If the first judgment result is negative, the corresponding sub-interval will not be adjusted.

4. The method according to claim 3, characterized in that, The step of adjusting the initial dictionary matrix based on the adjusted sub-intervals to obtain the target dictionary matrix includes: Based on the adjusted sub-intervals, the initial dictionary matrix is ​​adjusted to obtain the adjusted dictionary matrix; Set the value of k to 1; use the adjusted dictionary matrix as the initial data for the k-th iteration; Based on the signal data to be processed and the initial data of the k-th iteration, the time delay value of the k-th iteration is calculated; Determine whether the time delay value of the k-th iteration satisfies the first iteration termination condition, and obtain the second determination result; If the second judgment result is negative, the initial data of the kth iteration is adjusted to obtain the initial data of the (k+1)th iteration, the value of k is incremented by 1, and the step of calculating the time delay value of the kth iteration based on the signal data to be processed and the initial data of the kth iteration is returned. If the second judgment result is yes, then the iteration ends, and the initial data of the kth iteration is output as the target dictionary matrix.

5. The method according to claim 1, characterized in that, The step of iteratively optimizing the first and second hyperparameters using a deep unfolded network to obtain the target mean and target covariance of the time delay includes: Set the initial values ​​for the first hyperparameter and the second hyperparameter; Set the value of k to 1; use the initial values ​​of the first hyperparameter and the second hyperparameter as the initial data for the kth iteration; Based on the initial data of the k-th iteration, the mean and covariance of the k-th iteration are calculated. Based on the mean and covariance of the k-th iteration, the initial data of the k-th iteration is updated to obtain the initial data of the (k+1)-th iteration. Determine whether the initial data of the (k+1)th iteration satisfies the second iteration termination condition to obtain the third determination result; If the third judgment result is negative, then the initial data of the (k+1)th iteration is adjusted to obtain the initial data of the (k+2)th iteration, the value of k is incremented by 1, and the step of calculating the mean and covariance of the kth iteration based on the initial data of the kth iteration is returned. If the third judgment result is yes, the iteration ends, and the target mean and target covariance of the time delay are calculated based on the initial data of the k-th iteration.

6. The method according to claim 1, characterized in that, The process of processing the signal data to be processed and the target dictionary matrix to obtain the posterior distribution of the time delay includes: Based on the signal data to be processed and the target dictionary matrix, the likelihood probability distribution is calculated; Based on the automatic correlation decision method, the prior probability distribution of time delay is determined; Based on the likelihood probability distribution and the prior probability distribution, the marginal likelihood probability distribution of the time delay is calculated; Based on the likelihood probability distribution, the prior probability distribution, and the marginal likelihood probability distribution, the posterior distribution of the time delay is calculated.

7. A device for determining time delay, characterized in that, include: The initial delay result determination module is used to construct an initial dictionary matrix based on the signal data to be processed, and to calculate the initial delay result corresponding to the signal to be processed based on the signal data to be processed and the initial dictionary matrix. The dictionary matrix adjustment module is used to adjust the initial dictionary matrix based on the initial delay result to obtain the target dictionary matrix; The target delay result determination module is used to calculate the target delay result based on the signal data to be processed and the target dictionary matrix; The step of adjusting the initial dictionary matrix based on the initial delay result to obtain the target dictionary matrix includes: Calculate the residual energy value corresponding to the initial time delay result; Determine whether the residual energy value exceeds a first preset threshold. If so, divide the preset range of the time delay into multiple sub-intervals, calculate the residual energy value of each sub-interval, and adjust the initial dictionary matrix based on the residual energy value of each sub-interval to obtain the target dictionary matrix. If not, use the initial dictionary matrix as the target dictionary matrix. The calculation of the target time delay result based on the signal data to be processed and the target dictionary matrix includes: The signal data to be processed and the target dictionary matrix are processed to obtain the posterior distribution of the time delay; the posterior distribution of the time delay includes a first hyperparameter and a second hyperparameter; The first hyperparameter and the second hyperparameter are iteratively optimized by a deep unfolded network to obtain the target mean and target covariance of the time delay. Based on the target mean and the target covariance, the posterior distribution of the time delay is updated, and the target time delay result is determined based on the updated posterior distribution of the time delay.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the delay determination method as described in any one of claims 1-6.

Images