A radar link periodic error correction method, system, device and medium

Abstract

The present application relates to the technical field of radar error correction, and in particular to a radar link periodic error correction method, system, device and medium, comprising obtaining reference difference frequency information, performing phase unwrapping on the reference difference frequency information to obtain phase error information; performing an initial estimation on the phase error information based on a pre-constructed frequency estimation model and a frequency counting model to obtain first initial parameter information; constructing phase matching function information based on the first initial parameter information, analyzing the phase matching function information and the phase error information to obtain second initial parameter information; performing a first optimization process based on the first initial parameter information and the second initial parameter information to obtain coarse optimization result information, performing a second optimization process on the coarse optimization result information to obtain updated parameter information; and correcting the difference frequency information based on the updated parameter information to obtain correction result information, thereby more accurately performing periodic error correction and improving the application effect of the radar system in complex environments.

Description

Technical Field

[0001] The present application relates to the field of radar error correction technology, and in particular to a radar link periodic error correction method, system, device and medium. Background Art

[0002] Frequency-modulated continuous wave (FMCW) radar, due to its simple structure, low cost, and low probability of intercept, is widely used in fields such as close-range safety detection, autonomous driving, deformation monitoring, and vital sign monitoring. To achieve higher-precision target distance resolution, radar systems typically require an extremely large instantaneous bandwidth. This extremely large bandwidth places very high demands on the acquisition system, and analog demodulation technology has become an important means of solving this problem. However, this technology places extremely strict demands on the linear frequency modulation of the transmit and receive signals. Especially when the signal bandwidth is large, deviations in linearity can lead to severe blurring of target distance resolution and even inaccurate imaging. The transmit and receive links of radar systems are prone to introducing certain nonlinear errors. For example, common signal generation methods such as digital frequency synthesis (DDS) inherently have nonlinear characteristics. The influence of components such as frequency multipliers, mixers, and filters further reduces signal linearity. Current nonlinearity correction methods are divided into hardware and software. The hardware method minimizes FM nonlinearity by adjusting the signal generation circuit and can output high-linearity FM signals in real time. However, it is affected by environmental factors such as temperature and has poor stability. In comparison, the software method has greater versatility and adaptability and can perform nonlinear error correction on the sampled signal through digital signal processing technology.

[0003] Existing software correction methods primarily estimate and correct for transmit link nonlinearities, employing polynomial models to model phase errors. However, periodic errors are often caused by factors such as DDS spurious signals and clock spurious signals, and existing algorithms have yet to effectively address this issue. Fitting nonlinear parameters using techniques such as wavelet transforms and eigendecomposition still presents challenges such as high signal sampling rate requirements and performance degradation at low signal-to-noise ratios. Furthermore, existing periodic error correction methods lack accuracy in frequency estimation and amplitude and phase parameter optimization, making them incapable of meeting high-precision requirements in broadband radar applications. This limits the effectiveness of radar systems in complex environments, and these issues remain to be addressed. Summary of the Invention

[0004] In order to more accurately correct periodic errors and improve the application effect of radar systems in complex environments, this application provides a radar link periodic error correction method, system, device, and medium, which adopt the following technical solutions:

[0005] In a first aspect, the present application provides a radar link periodic error correction method, comprising:

[0006] Obtain reference difference frequency information, perform phase unwrapping on the reference difference frequency information, and obtain phase error information;

[0007] Performing an initial estimation on the phase error information according to a pre-built frequency estimation model and a frequency counting model to obtain first initial parameter information;

[0008] constructing phase matching function information according to the first initial parameter information, and analyzing the phase matching function information and the phase error information to obtain second initial parameter information;

[0009] Performing a first optimization process on the first initial parameter information and the second initial parameter information to obtain rough optimization result information, and performing a second optimization process on the rough optimization result information to obtain updated parameter information;

[0010] The frequency difference information is corrected according to the updated parameter information to obtain correction result information.

[0011] Preferably, the specific steps of performing an initial estimation on the phase error information based on the pre-built frequency estimation model and frequency counting model to obtain the first initial parameter information are:

[0012] The frequency value of the phase error information is estimated according to the pre-built frequency estimation model to obtain the initial frequency value information.

[0013] Preferably, the specific steps of performing an initial estimation on the phase error information based on the pre-built frequency estimation model and frequency counting model to obtain the first initial parameter information are:

[0014] According to the pre-built frequency counting model, the frequency number of the phase error information is estimated to obtain the frequency number information.

[0015] Preferably, the phase matching function information is:

[0016]

[0017] in, is the initial frequency value information, and K is the frequency number information.

[0018] Preferably, the specific steps of analyzing the phase matching function information and the phase error information to obtain the second initial parameter information are:

[0019] Multiplying the phase matching function information and the phase error information to obtain DC component information;

[0020] Performing maximum value analysis on DC component information to obtain DC component maximum value information;

[0021] Phase analysis is performed based on the maximum value information of the DC component to obtain initial phase parameter information.

[0022] Preferably, the specific step of analyzing the phase matching function information and the phase error information to obtain the second initial parameter information further includes:

[0023] Multiplying the phase matching function information and the phase error information to obtain DC component information;

[0024] Performing maximum value analysis on DC component information to obtain DC component maximum value information;

[0025] Amplitude analysis is performed based on the maximum value information of the DC component to obtain initial amplitude parameter information.

[0026] Preferably, it also includes:

[0027] The first optimization process is a rough optimization performed by the Adam algorithm;

[0028] The second optimization process is to perform fine optimization using the L-BFGS algorithm.

[0029] In a second aspect, the present application provides a radar link periodic error correction system, comprising:

[0030] An information acquisition module is used to obtain reference difference frequency information, perform phase unwrapping on the reference difference frequency information, and obtain phase error information;

[0031] A first initial parameter analysis module is used to perform an initial estimation on the phase error information according to a pre-built frequency estimation model and a frequency counting model to obtain first initial parameter information;

[0032] A second initial parameter analysis module is used to construct phase matching function information based on the first initial parameter information, and analyze the phase matching function information and the phase error information to obtain second initial parameter information;

[0033] An update parameter analysis module is configured to perform a first optimization process based on the first initial parameter information and the second initial parameter information to obtain rough optimization result information, and perform a second optimization process on the rough optimization result information to obtain update parameter information;

[0034] The correction module is used to correct the frequency difference information according to the updated parameter information to obtain correction result information.

[0035] Preferably, the first initial parameter analysis module performs an initial estimation on the phase error information according to a pre-built frequency estimation model and a frequency counting model, and obtains the first initial parameter information in the following specific steps:

[0036] The frequency value of the phase error information is estimated according to the pre-built frequency estimation model to obtain the initial frequency value information.

[0037] Preferably, the first initial parameter analysis module performs an initial estimation on the phase error information according to a pre-built frequency estimation model and a frequency counting model, and obtains the first initial parameter information in the following specific steps:

[0038] According to the pre-built frequency counting model, the frequency number of the phase error information is estimated to obtain the frequency number information.

[0039] Preferably, the phase matching function information of the second initial parameter analysis module is:

[0040]

[0041] in, is the initial frequency value information, and K is the frequency number information.

[0042] Preferably, the specific steps of the second initial parameter analysis module analyzing the phase matching function information and the phase error information to obtain the second initial parameter information are:

[0043] Multiplying the phase matching function information and the phase error information to obtain DC component information;

[0044] Performing maximum value analysis on DC component information to obtain DC component maximum value information;

[0045] Phase analysis is performed based on the maximum value information of the DC component to obtain initial phase parameter information.

[0046] Preferably, the specific step of the second initial parameter analysis module analyzing the phase matching function information and the phase error information to obtain the second initial parameter information further includes:

[0047] Multiplying the phase matching function information and the phase error information to obtain DC component information;

[0048] Performing maximum value analysis on DC component information to obtain DC component maximum value information;

[0049] Amplitude analysis is performed based on the maximum value information of the DC component to obtain initial amplitude parameter information.

[0050] Preferably, it also includes:

[0051] The first optimization process of the update parameter analysis module is to perform rough optimization using the Adam algorithm;

[0052] The second optimization process of the update parameter analysis module is to perform fine optimization through the L-BFGS algorithm.

[0053] In a third aspect, the present application provides a radar link periodic error correction device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute the radar link periodic error correction method as described above.

[0054] In a fourth aspect, the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, wherein the computer program is configured to execute the radar link periodic error correction method as described above when running.

[0055] In summary, compared with the prior art, the technical solution provided by this application has at least the following beneficial effects:

[0056] The present application obtains reference difference frequency information containing periodic phase error, performs phase unwrapping on the reference difference frequency information to obtain phase error information, uses a frequency estimation model and a frequency counting model to make a preliminary estimate of the periodic phase error, thereby obtaining first initial parameter information of the periodic phase error, adopts a phase matching method based on the first initial parameter information to obtain second initial parameter information of the periodic phase error, uses an optimization algorithm to process the first initial parameter information and the second initial parameter information, further optimizes to obtain updated parameter information, and finally corrects the difference frequency information based on the updated parameter information, effectively eliminating the periodic phase error in the difference frequency information, performing periodic error correction more accurately, and improving the application effect of the radar system in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 This is a flow chart of a radar link periodic error correction method described in an embodiment of the present application.

[0058] Figure 2 This is a module diagram of a radar link periodic error correction system described in an embodiment of the present application.

[0059] Description of reference numerals:

[0060] 1. Information acquisition module; 2. First initial parameter analysis module; 3. Second initial parameter analysis module; 4. Update parameter analysis module; 5. Correction module. DETAILED DESCRIPTION

[0061] The following combination Figure 1-Figure 2 To further explain the present application in detail, the terms used in the embodiments of the present application are only for the purpose of describing specific embodiments and are not intended to be limiting.

[0062] Reference Figure 1 The present application relates to a radar link periodic error correction method, specifically comprising:

[0063] Obtain reference difference frequency information, perform phase unwrapping on the reference difference frequency information, and obtain phase error information;

[0064] Performing an initial estimation on the phase error information according to a pre-built frequency estimation model and a frequency counting model to obtain first initial parameter information;

[0065] constructing phase matching function information according to the first initial parameter information, and analyzing the phase matching function information and the phase error information to obtain second initial parameter information;

[0066] Performing a first optimization process on the first initial parameter information and the second initial parameter information to obtain rough optimization result information, and performing a second optimization process on the rough optimization result information to obtain updated parameter information;

[0067] The frequency difference information is corrected according to the updated parameter information to obtain correction result information.

[0068] Specifically, the periodic phase error estimation and correction of the present application is based on periodic phase error modeling and analysis. First, the difference frequency error model and its periodic phase distortion analysis are completed, and then the periodic phase error estimation and correction process is carried out.

[0069] The modeling and analysis of periodic phase error is as follows: first, the difference frequency signal error modeling is performed. The ideal FMCW linear frequency modulation continuous wave RF signal model is expressed as:

[0070]

[0071] Among them, f c is the carrier frequency, K r is the frequency modulation slope, which is defined as the ratio of bandwidth to modulation period. t is time. The received signal is the frequency modulation signal after a time delay τ of the transmitted signal, which is expressed as:

[0072]

[0073] in, is the signal delay, R is the target distance, and c is the speed of light.

[0074] After mixing the received signal and the transmitted signal, namely Dechirp, a difference frequency signal with nonlinear error can be obtained, specifically:

[0075]

[0076] Where Kτt represents the target's beat frequency, is the residual video phase (RVP). In near-field imaging, the time delay is very small, so the RVP term will be ignored in subsequent processing. This application further models the phase error of the difference frequency signal as the sum of a polynomial phase set and a sine function set:

[0077]

[0078] Where L is the polynomial error order, p is the polynomial coefficient. M is the number of periodic phase error sinusoidal signals, a k 、ω k and φ k are the amplitude, angular frequency, and phase of the periodic phase error, respectively. Therefore, the difference frequency signal model containing nonlinear phase error is:

[0079]

[0080] τ ref is the time delay of the reference target. After removing the polynomial phase error, the difference frequency signal is:

[0081]

[0082] This application performs periodic phase distortion analysis. The frequency modulation linearity of the LFM signal is usually defined as Performance Analysis of Frequency Sweep Nonlinearities in LFM Radars1-4:

[0083]

[0084] Where, η represents the frequency modulation linearity of the LFM signal, B is the frequency modulation bandwidth of the LFM signal, and f e (t) Frequency shift.

[0085] The periodic phase error is then analyzed as follows: For the periodic frequency offset, the Fourier analysis point of view is used to decompose it into the sum of infinite simple harmonics. For the sake of simplicity of analysis, a single frequency frequency difference is introduced:

[0086] f e (t) = ηBcos(2πf m t)

[0087] f m is the frequency of the nonlinear single frequency deviation, so the LFM frequency domain function containing periodic phase error is:

[0088] f(t)=f c +K r t+ηBcos(2πf m t)fc is the carrier frequency, K r is the frequency modulation slope, K r =B / T, where T is the pulse width. Then the LFM phase signal with periodic frequency deviation distortion is:

[0089]

[0090] make Then the difference frequency signal with periodic phase error is:

[0091] s if (t)=exp{j2π[f c τ+K r tτ+βsin(2πf m t)]}

[0092] The ideal FMCW signal output signal after passing through the matched filter is s o (t), then when there is single-frequency periodic distortion, the output signal after matched filtering is:

[0093]

[0094] Among them J n (β) is the nth-order Bessel function of the first kind, E nv (·) is the de-envelope function.

[0095] It can be seen that when there is a periodic phase error, the signal output after pulse compression is mainly composed of the main signal with an amplitude of J0(β) and the main signal with an amplitude of J n The time interval and frequency interval between the paired echo and the main signal are ±nf m / K r ±nf m / 2.

[0096] Based on the above derivation, this application draws the following conclusions: the impact of periodic phase error on radar performance is mainly caused by paired echoes; periodic phase error will affect the detection effect of weak targets near the main lobe and distant targets; periodic phase error will not affect the main lobe width and main lobe position.

[0097] The present application obtains reference difference frequency information containing periodic phase error, performs phase unwrapping on the reference difference frequency information to obtain phase error information, uses a frequency estimation model and a frequency counting model to make a preliminary estimate of the periodic phase error, thereby obtaining first initial parameter information of the periodic phase error, adopts a phase matching method based on the first initial parameter information to obtain second initial parameter information of the periodic phase error, uses an optimization algorithm to process the first initial parameter information and the second initial parameter information, further optimizes to obtain updated parameter information, and finally corrects the difference frequency information based on the updated parameter information, effectively eliminating the periodic phase error in the difference frequency information, performing periodic error correction more accurately, and improving the application effect of the radar system in complex environments.

[0098] Among them, the error estimation method of the present application is a periodic error estimation method based on a deep network model, which performs polynomial error correction to obtain the phase signal of the difference frequency signal; obtains the initial value of the periodic error through a deep learning model and a phase matching method; uses the ADAM and LBFGS joint algorithm for optimization; uses the estimated parameters to perform phase error correction through Fourier matching, and finally obtains the corrected difference frequency signal.

[0099] As one implementation method, the specific steps of obtaining reference difference frequency information, performing phase unwrapping on the reference difference frequency information, and obtaining phase error information are as follows:

[0100] After the polynomial error correction, the signal containing periodic error is removed from the polynomial phase:

[0101]

[0102] Take this signal and perform phase unwrapping to obtain:

[0103]

[0104] Further simplifying this signal yields:

[0105]

[0106] Where a k =2πA k τ ref ζ=2πf c τ ref , considering ζ as a constant will not affect the estimation of subsequent sinusoidal parameters. ζ can be eliminated by removing DC before fitting.

[0107] As one implementation method, the specific steps of performing an initial estimation on the phase error information based on the pre-built frequency estimation model and frequency counting model to obtain the first initial parameter information are as follows:

[0108] The frequency value of the phase error information is estimated according to the pre-built frequency estimation model to obtain the initial frequency value information.

[0109] Specifically, the first initial parameter information in the embodiment of the present application includes initial frequency value information. A model suitable for estimating periodic phase errors, namely a frequency estimation module, is trained using simulation data. This module converts the input signal into a frequency representation to extract the frequency characteristics of the signal. This module includes an input layer, multiple convolutional layers, and an output layer, and uses a narrow Gaussian kernel to generate an ideal frequency representation.

[0110] The input layer of the frequency estimation model undergoes a preliminary feature transformation of the input signal through a linear layer. This linear layer projects the input signal into a higher-dimensional internal representation space, providing a rich feature representation for subsequent convolution operations. By leveraging the powerful mapping capabilities of the fully connected layer, the initial features of the signal can be fully extracted.

[0111] The convolutional layer of the frequency estimation model processes the output of the input layer through multiple one-dimensional convolutional layers. The convolutional layer stems from the successful application of convolutional neural networks in image processing. Its main advantage is its ability to effectively capture local features in the signal. Each convolutional layer uses multiple filters, and the convolution operation is followed by a batch normalization layer and a Reluctant Unit (ReLU) activation function. The convolution kernel slides over the signal to generate a feature map that captures frequency information at different scales, extracting the local frequency features of the signal through the convolution operation. The batch normalization layer helps accelerate the training process and stabilizes the model training, preventing gradient vanishing or exploding. This structure enables the convolutional layer to extract rich frequency features while ensuring computational efficiency, and to refine the high-level features of the signal layer by layer through multi-layer stacking.

[0112] The output layer of the frequency estimation model uses a deconvolution layer, also known as a transposed convolution layer, to upsample the internal representation to the desired frequency representation size. The deconvolution layer is a special operation in convolutional neural networks that acts in the opposite way to the convolution operation. Convolution is typically used for downsampling, which involves extracting features and reducing the data size through a sliding window. Deconvolution, on the other hand, is used for upsampling, which involves restoring the data size through interpolation and kernel expansion. The upsampling in the deconvolution layer allows the output frequency representation to retain greater detail, facilitating subsequent frequency estimation and counting tasks. High-resolution frequency representations can more accurately capture subtle frequency variations in the signal, thereby improving the accuracy of frequency estimation.

[0113] The embodiment of the present application uses a narrow-band Gaussian kernel to generate ideal frequency information when training the model, specifically: generating a frequency representation, first performing a Fourier transform on the input signal to obtain a frequency domain representation; convolution operation, then convolving the frequency domain representation with the narrow Gaussian kernel to generate an ideal frequency representation. This step is usually performed in the target generation stage of model training to generate a target frequency representation for training; smoothing processing, through the convolution operation of the narrow-band Gaussian kernel, each frequency component forms a peak in the frequency representation, highlighting the significance of the frequency component and suppressing noise and other interference components.

[0114] The loss function uses the mean squared error loss function, MSE, to measure the difference between the predicted frequency representation and the true frequency representation:

[0115]

[0116] The MSE loss function can effectively reflect the average deviation between the predicted value and the true value. Through the structural design of the embodiment of the present application, the frequency representation module can effectively extract frequency features from the input signal and generate a high-resolution frequency representation. This architectural design can not only capture the key frequency components in the signal, but also adapt to various noise and interference conditions, thereby improving the accuracy and robustness of frequency estimation. The application of a narrow Gaussian kernel further enhances the quality of the frequency representation, enabling the frequency estimation task to perform well in high-noise and complex signal environments.

[0117] As one implementation method, the specific steps of performing an initial estimation on the phase error information based on the pre-built frequency estimation model and frequency counting model to obtain the first initial parameter information are as follows:

[0118] According to the pre-built frequency counting model, the frequency number of the phase error information is estimated to obtain the frequency number information.

[0119] Specifically, the first initial parameter information in the embodiment of the present application also includes frequency count information. The frequency counting module in the embodiment of the present application is similar to the frequency estimation module. The frequency counting module extracts the number of frequencies in the signal from the frequency representation. This module also includes an input layer, multiple convolutional layers, and an output layer. The input layer and convolutional layers are the same as those in the frequency estimation module. The output of the frequency counting module uses a linear layer to convert the internal representation into the final frequency count result.

[0120] When the number of frequencies is a discrete value and the possible number of frequencies is finite and known in advance, the model adjusts the loss function to the cross entropy loss function:

[0121]

[0122] As one implementation method, the phase matching function information is:

[0123]

[0124] in, is the initial frequency value information, and K is the frequency number information.

[0125] Specifically, the embodiment of the present application obtains the initial frequency value through frequency estimation and frequency counting model And the number of frequencies K. Use the estimated frequency parameters to construct the phase matching function, and multiply the phase matching function with the phase error signal to obtain:

[0126]

[0127] As one implementation manner, the specific steps of analyzing the phase matching function information and the phase error information to obtain the second initial parameter information are:

[0128] Multiplying the phase matching function information and the phase error information to obtain DC component information;

[0129] Performing maximum value analysis on DC component information to obtain DC component maximum value information;

[0130] Phase analysis is performed based on the maximum value information of the DC component to obtain initial phase parameter information.

[0131] Specifically, when At this time, the DC component is By traversing θ at (-π, π), when the DC component reaches its maximum value, the phase parameter of the frequency component can be obtained.

[0132] As one implementation manner, the specific step of analyzing the phase matching function information and the phase error information to obtain the second initial parameter information further includes:

[0133] Multiplying the phase matching function information and the phase error information to obtain DC component information;

[0134] Performing maximum value analysis on DC component information to obtain DC component maximum value information;

[0135] Amplitude analysis is performed based on the maximum value information of the DC component to obtain initial amplitude parameter information.

[0136] Specifically, in the embodiment of the present application, when the DC component takes the maximum value, the phase parameter of the frequency component is obtained, and half of the DC component is the amplitude parameter of the frequency component. Therefore, only one-dimensional search is required to obtain Let k = k + 1 and update the error phase:

[0137]

[0138] The embodiment of the present application repeats the above phase matching process to obtain the parameter values ​​of K sinusoidal components and obtain the initial parameter values Initial value used for subsequent optimization algorithms.

[0139] As one implementation method, it also includes:

[0140] The first optimization process is a rough optimization performed by the Adam algorithm;

[0141] The second optimization process is to perform fine optimization using the L-BFGS algorithm.

[0142] Specifically, the Adam optimization algorithm used in this embodiment is more efficient and robust than traditional gradient descent methods such as SGD during optimization, achieving better performance with fewer iterations. However, it cannot always find the global optimal solution. Therefore, it is necessary to combine it with other optimizers to achieve better optimization results. Therefore, this embodiment uses L-BFGS for fine optimization.

[0143] Among them, the Adam loss function is the mean square error:

[0144]

[0145]

[0146] Among them, θ represents the optimization parameter a i ,ω i ,φ i ,y ture is the input data to be fitted.

[0147] In this embodiment of the application, a K value is initially set. If there is no historical experience information, it starts from 1 and there will be subsequent K value training. The initial sine parameters are given in the previous section and are recorded as

[0148] Initialize the first-order moment estimation momentum variable m0=0 and the second-order moment estimation root mean square variable v0=0 as zero vectors, set the learning rate α, the first-order momentum hyperparameter β1, the second-order momentum hyperparameter β2, β1 is usually close to 1, such as 0.9, β2 is usually close to 1, such as 0.999, and a small constant ε=10 to prevent division by zero errors -8 .

[0149] At each iteration, the gradient of the current parameters is calculated:

[0150]

[0151] Among them, θ represents the three parameters of amplitude, frequency and phase a i ,ω i ,φ i , so the gradient of each parameter is:

[0152]

[0153]

[0154]

[0155] Update the first-order moment estimate, i.e., the momentum variable:

[0156] m t =β1m t-1 +(1-β1)g t

[0157] where m t Is the momentum variable at the current moment, representing the exponentially weighted average of the gradient. Update the second-order moment estimate, that is, the root mean square variable:

[0158]

[0159] where v t is the root mean square variable at the current moment, which represents the exponentially weighted average of the square of the gradient. Next, the bias correction is calculated. Since m t and v t Initialized to zero vector, so there is deviation in the initial stage, the following deviation correction is performed:

[0160]

[0161]

[0162] Then use the corrected moment estimates to update the parameters:

[0163]

[0164] Among them, θ represents the optimization parameter a i ,ω i ,φ i , the coarsely optimized periodic phase error parameter can be obtained, and then this parameter is used for local optimization to obtain a more accurate error parameter.

[0165] In the second optimization process of the embodiment of the present application, that is, in the second stage of optimization, the L-BFGS optimization algorithm is used to further fine-tune the model parameters and improve the fitting accuracy. L-BFGS is a quasi-Newton method that deals with optimization problems requiring high-precision convergence. It approximates the inverse of the Hessian matrix through limited memory, making it more efficient in large-scale optimization problems. In the embodiment of the present application, the L-BFGS algorithm is used to optimize the amplitude, frequency, and phase parameters so that the generated signal matches the real signal as closely as possible. Through gradient calculation, historical record update, search direction calculation, and parameter update, the L-BFGS algorithm gradually optimizes the model parameters to achieve accurate fitting of the signal. Specifically:

[0166] The initialized fitting parameter θ0 is the final result of Adam optimization Initialize a set of history records {S j , G j} is used to store gradients and position vectors, and the immediate record size is set to 20.

[0167] The method for iteratively updating the gradient g in this embodiment of the application is similar to Adam and will not be repeated here. Then, the position vector difference and gradient difference are updated to record the updated historical information, and the inverse of the Hessian matrix is ​​approximately calculated based on the historical information:

[0168] S i =θ j+1 -θ j

[0169] G j =g j+1 -g j

[0170] Then calculate the auxiliary variable ρ j :

[0171]

[0172] Initialization vector q = g j The first loop starts from the most recent history and traverses forward, computing the scalar η j And update the vector q:

[0173]

[0174] q=q-η j G j

[0175] Then calculate the scaling factor γ of the initial matrix j And initialize the vector r:

[0176]

[0177] r = γ j q

[0178] The second loop starts from the furthest historical record and traverses backwards, computing the scalar χ j And update the vector r:

[0179]

[0180] r=r+S j (η j -χ j )

[0181] From the above we get the search direction p j , determine the direction of parameter update so that the loss function decreases fastest along this direction:

[0182] p j =-r

[0183] Finally, update the parameters to get:

[0184] θ j+1 =θ j +χ j p j

[0185] If the norm of the gradient ||g j || is less than the set tolerance value ζ, or reaches the maximum number of iterations, the iteration is stopped and the precise optimization process is completed.

[0186] In the training process of the multi-sinusoidal signal model of the embodiment of the present application, the time and signal data are first read, and the data is preprocessed to remove the boundary data points. Then, the initial amplitude, frequency and phase parameters are obtained through the neural network model and phase matching to initialize the neural network model. The model uses the mean square error MSE as the loss function and is preliminarily trained by the Adam optimizer, and the optimal parameter settings are recorded in the process. The model parameters are further optimized by the L-BFGS optimizer, and the fitting accuracy is improved by the quasi-Newton method. In order to further fine-tune the parameters, the least squares method is used for final optimization, and the optimal amplitude, frequency and phase parameters are finally returned for generating the fitting signal, and the results are evaluated and visualized. The comprehensive training process ensures efficient optimization and accurate fitting of the model.

[0187] As one implementation method, the reference difference frequency information is corrected according to the updated parameter information to obtain the correction result information. The step is to use a more widely applicable matched Fourier transform to achieve multi-target nonlinear correction. When a nonlinear analytical expression is obtained, the multi-target can be focused. The general form of the specific MFT is:

[0188]

[0189] The multi-target difference frequency signal containing sinusoidal periodic phase error is expressed as:

[0190]

[0191] Where Q is the target number; τ q is the echo delay of the qth target;

[0192] The Fourier transform is:

[0193]

[0194] The embodiment of this application Then s if The MFT of (t) is:

[0195]

[0196] Where ψ′(t) is the derivative of ψ(t), and its expression is:

[0197]

[0198] Point targets at different distances are focused as an ideal sinc function, with the target frequency being f p =K r τ p The real distance of the target is reflected and the nonlinearity of the system is corrected.

[0199] The period amplitude of the new time axis ψ(t) is:

[0200]

[0201] From the time axis t to the new time axis ψ(t) is the resampling process, the intermediate frequency signal s if The MFT of (t) is equivalent to the Fourier transform on the new time axis ψ(t), and the one-dimensional range image compression result on the new time axis ψ(t) is the compression result after the periodic phase error correction.

[0202] Reference Figure 2 , an embodiment of the present application provides a radar link periodic error correction system, comprising:

[0203] An information acquisition module is used to obtain reference difference frequency information, perform phase unwrapping on the reference difference frequency information, and obtain phase error information;

[0204] A first initial parameter analysis module is used to perform an initial estimation on the phase error information according to a pre-built frequency estimation model and a frequency counting model to obtain first initial parameter information;

[0205] A second initial parameter analysis module is used to construct phase matching function information based on the first initial parameter information, and analyze the phase matching function information and the phase error information to obtain second initial parameter information;

[0206] An update parameter analysis module is configured to perform a first optimization process based on the first initial parameter information and the second initial parameter information to obtain rough optimization result information, and perform a second optimization process on the rough optimization result information to obtain update parameter information;

[0207] The correction module is used to correct the frequency difference information according to the updated parameter information to obtain correction result information.

[0208] As one implementation manner, the first initial parameter analysis module performs an initial estimation on the phase error information based on a pre-built frequency estimation model and a frequency counting model, and the specific steps for obtaining the first initial parameter information are as follows:

[0209] The frequency value of the phase error information is estimated according to the pre-built frequency estimation model to obtain the initial frequency value information.

[0210] As one implementation manner, the first initial parameter analysis module performs an initial estimation on the phase error information based on a pre-built frequency estimation model and a frequency counting model, and the specific steps for obtaining the first initial parameter information are as follows:

[0211] According to the pre-built frequency counting model, the frequency number of the phase error information is estimated to obtain the frequency number information.

[0212] As one implementation method, the phase matching function information of the second initial parameter analysis module is:

[0213]

[0214] in, is the initial frequency value information, and K is the frequency number information.

[0215] As one implementation manner, the specific steps of the second initial parameter analysis module analyzing the phase matching function information and the phase error information to obtain the second initial parameter information are as follows:

[0216] Multiplying the phase matching function information and the phase error information to obtain DC component information;

[0217] Performing maximum value analysis on DC component information to obtain DC component maximum value information;

[0218] Phase analysis is performed based on the maximum value information of the DC component to obtain initial phase parameter information.

[0219] As one implementation manner, the specific step of the second initial parameter analysis module analyzing the phase matching function information and the phase error information to obtain the second initial parameter information further includes:

[0220] Multiplying the phase matching function information and the phase error information to obtain DC component information;

[0221] Performing maximum value analysis on DC component information to obtain DC component maximum value information;

[0222] Amplitude analysis is performed based on the maximum value information of the DC component to obtain initial amplitude parameter information.

[0223] As one implementation method, it also includes:

[0224] The first optimization process of the update parameter analysis module is to perform rough optimization using the Adam algorithm;

[0225] The second optimization process of the update parameter analysis module is to perform fine optimization through the L-BFGS algorithm.

[0226] An embodiment of the present application provides a radar link periodic error correction device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to perform the radar link periodic error correction method as described above.

[0227] An embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, wherein the computer program is configured to execute the radar link periodic error correction method as described above when running.

[0228] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices and products can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0229] In several embodiments provided in this application, it should be understood that the disclosed methods, systems, devices, and program products may be implemented in other ways.

[0230] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0231] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A radar link periodic error correction method, characterized in that: include: Obtain reference difference frequency information, perform phase unwrapping on the reference difference frequency information, and obtain phase error information; Performing an initial estimation on the phase error information according to a pre-built frequency estimation model and a frequency counting model to obtain first initial parameter information; constructing phase matching function information according to the first initial parameter information, and analyzing the phase matching function information and the phase error information to obtain second initial parameter information; Performing a first optimization process on the first initial parameter information and the second initial parameter information to obtain rough optimization result information, and performing a second optimization process on the rough optimization result information to obtain updated parameter information; The frequency difference information is corrected according to the updated parameter information to obtain correction result information.

2. The radar link periodic error correction method according to claim 1, characterized in that: The specific steps of performing an initial estimation on the phase error information based on the pre-built frequency estimation model and the frequency counting model to obtain the first initial parameter information are: The frequency value of the phase error information is estimated according to the pre-built frequency estimation model to obtain the initial frequency value information.

3. The radar link periodic error correction method according to claim 2, characterized in that: The specific steps of performing an initial estimation on the phase error information based on the pre-built frequency estimation model and the frequency counting model to obtain the first initial parameter information are: According to the pre-built frequency counting model, the frequency number of the phase error information is estimated to obtain the frequency number information.

4. The radar link periodic error correction method according to claim 3, characterized in that: The phase matching function information is: in, is the initial frequency value information, K is the frequency number information, and t is the time.

5. The radar link periodic error correction method according to claim 4, characterized in that: The specific steps of analyzing the phase matching function information and the phase error information to obtain the second initial parameter information are: Multiplying the phase matching function information and the phase error information to obtain DC component information; Performing maximum value analysis on DC component information to obtain DC component maximum value information; Phase analysis is performed based on the maximum value information of the DC component to obtain initial phase parameter information.

6. The radar link periodic error correction method according to claim 5, characterized in that: The specific step of analyzing the phase matching function information and the phase error information to obtain the second initial parameter information further includes: Multiplying the phase matching function information and the phase error information to obtain DC component information; Performing maximum value analysis on DC component information to obtain DC component maximum value information; Amplitude analysis is performed based on the maximum value information of the DC component to obtain initial amplitude parameter information.

7. The radar link periodic error correction method according to claim 1, characterized in that: Also includes: The first optimization process is a rough optimization performed by the Adam algorithm; The second optimization process is to perform fine optimization using the L-BFGS algorithm.

8. A radar link periodic error correction system, characterized in that: include: An information acquisition module is used to obtain reference difference frequency information, perform phase unwrapping on the reference difference frequency information, and obtain phase error information; A first initial parameter analysis module is used to perform an initial estimation on the phase error information according to a pre-built frequency estimation model and a frequency counting model to obtain first initial parameter information; A second initial parameter analysis module is used to construct phase matching function information based on the first initial parameter information, and analyze the phase matching function information and the phase error information to obtain second initial parameter information; An update parameter analysis module is configured to perform a first optimization process based on the first initial parameter information and the second initial parameter information to obtain rough optimization result information, and perform a second optimization process on the rough optimization result information to obtain update parameter information; The correction module is used to correct the frequency difference information according to the updated parameter information to obtain correction result information.

9. The radar link periodic error correction system according to claim 8, characterized in that: The first initial parameter analysis module performs an initial estimation on the phase error information based on the pre-built frequency estimation model and frequency counting model, and obtains the first initial parameter information in the following specific steps: The frequency value of the phase error information is estimated according to the pre-built frequency estimation model to obtain the initial frequency value information.

10. The radar link periodic error correction system according to claim 9, characterized in that: The first initial parameter analysis module performs an initial estimation on the phase error information based on the pre-built frequency estimation model and frequency counting model, and obtains the first initial parameter information in the following specific steps: According to the pre-built frequency counting model, the frequency number of the phase error information is estimated to obtain the frequency number information.

11. The radar link periodic error correction system according to claim 10, characterized in that: The phase matching function information of the second initial parameter analysis module is: in, is the initial frequency value information, K is the frequency number information, and t is the time.

12. The radar link periodic error correction system according to claim 11, characterized in that: The specific steps of the second initial parameter analysis module analyzing the phase matching function information and the phase error information to obtain the second initial parameter information are: Multiplying the phase matching function information and the phase error information to obtain DC component information; Performing maximum value analysis on DC component information to obtain DC component maximum value information; Phase analysis is performed based on the maximum value information of the DC component to obtain initial phase parameter information.

13. The radar link periodic error correction system according to claim 12, characterized in that: The specific step of the second initial parameter analysis module analyzing the phase matching function information and the phase error information to obtain the second initial parameter information further includes: Multiplying the phase matching function information and the phase error information to obtain DC component information; Performing maximum value analysis on DC component information to obtain DC component maximum value information; Amplitude analysis is performed based on the maximum value information of the DC component to obtain initial amplitude parameter information.

14. The radar link periodic error correction system according to claim 8, characterized in that: Also includes: The first optimization process of the update parameter analysis module is to perform rough optimization using the Adam algorithm; The second optimization process of the update parameter analysis module is to perform fine optimization through the L-BFGS algorithm.

15. A radar link periodic error correction device, characterized in that: The invention comprises a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to perform the radar link periodic error correction method according to any one of claims 1 to 7.

16. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, wherein the computer program is configured to execute the radar link periodic error correction method according to any one of claims 1 to 7 when running.

Images