Meter-wave radar low-elevation-angle multipath-removing high-precision angle measurement method based on characteristic game

Abstract

The invention discloses a meter-wave radar low-elevation-angle multipath-removing high-precision angle measurement method based on a characteristic game, and relates to the technical field of radar signal processing. The invention provides a meter-wave radar low-elevation-angle multipath-removing high-precision angle measurement method based on a characteristic game, and the method comprises the steps: building a receiving signal model of an antenna array according to an actual position reflection condition; constructing a generative adversarial network model; optimizing the generative adversarial network model based on a feature game and multipath suppression, and enhancing the signal features of target echoes; and based on the enhanced signal, carrying out angle measurement by using a classical positioning estimation algorithm. Therefore, by adopting the meter-wave radar low-elevation-angle multipath-removing high-precision angle measurement method based on the characteristic game, the separation of the target echo signal and the multipath signal can be realized, and the angle measurement precision of the meter-wave radar can be improved under the condition that the signal-to-noise ratio is matched or not completely matched.

Description

Technical Field

[0001] The present invention relates to the technical field of radar signal processing, and in particular to a high-precision angle measurement method for low-elevation-angle multipath removal of meter-wave radar based on feature game. Background Art

[0002] Meter-wave radar occupies an important position in modern radar technology, especially in the military and aviation fields. Due to its unique long wavelength characteristics, meter-wave radar has significant advantages in combating stealth targets and penetrating complex terrain.

[0003] However, in low-elevation angle measurement applications, the multipath interference problem faced by meter-wave radar is particularly prominent. The multipath effect is more obvious at low elevation angles, resulting in multiple path interference when receiving signals, which seriously affects the radar's angle measurement accuracy.

[0004] At present, most traditional multipath mitigation methods rely on supervised signal processing technology, which requires a large amount of labeled data for model training. In addition, obtaining high-quality radar labeled data is often costly and difficult in practical applications, which to some extent limits the popularization and application of these methods.

[0005] Unsupervised multipath suppression technology aims to automatically identify and suppress the interference of multipath signals without pre-labeled data by exploring the characteristics and distribution patterns of the signal itself, and has higher flexibility and applicability. Therefore, it is urgent to establish a high-precision angle measurement method for unsupervised multipath suppression to provide a reference for the widespread application of meter-wave radar. Summary of the invention

[0006] The purpose of the present invention is to provide a high-precision angle measurement method for low-elevation-angle multipath elimination of meter-wave radar based on feature game, which can solve the problem of suppressing multipath interference faced by meter-wave radar and improve the angle measurement accuracy of meter-wave radar.

[0007] To achieve the above object, the present invention provides a high-precision angle measurement method for low-elevation-angle multipath removal of meter-wave radar based on feature game, comprising the following steps:

[0008] S1. Establish a receiving signal model of the antenna array according to the actual position reflection conditions;

[0009] S2, build a generative adversarial network model;

[0010] S3. Based on feature game and multipath suppression, the generative adversarial network model is optimized to enhance the target echo signal;

[0011] S4. Based on the enhanced signal, the angle is measured using the classical positioning estimation algorithm.

[0012] Preferably, in step S1, the expression of the received signal model is as follows:

[0013] y(t)=(a(θ d )+ρexp(-j2πΔR / λ)a(θ r ))s(t)+n(t),t=1,2,...,T

[0014] Where t is the number of snapshots, y(t) is the tth snapshot data, and a(θ d ) is the direct waveguide vector, a(θ r ) is the multipath steering vector, ρ is the complex reflection coefficient, s(t) is the unit signal complex envelope, and n(t) is the noise.

[0015] Preferably, in step S2, the generative adversarial network model includes a complex domain generator and a complex domain discriminator, as follows:

[0016] The complex domain generator is used to extract the direct wave signal from the low signal-to-noise ratio echo signal and separate it from the multipath signal; the complex domain discriminator is used to explore the potential high-dimensional nonlinear features in the real high signal-to-noise ratio signal, and distinguish the high signal-to-noise ratio signal generated by the complex domain generator from the real high signal-to-noise ratio signal by analyzing, comparing and integrating the high-dimensional nonlinear features.

[0017] Preferably, step S3 includes optimizing the complex domain discriminator and the complex domain generator using an unsupervised multipath suppression method, and the objective function is:

[0018]

[0019] In the formula, G is the complex domain generator, D is the complex domain discriminator, and s t 、s t ' are real signal samples and generated signal samples respectively, D s , D s ' are the real signal space and the pseudo signal space respectively.

[0020] Preferably, the optimization of the generative adversarial network model includes constraining the characteristics and power of the generated signal, as follows:

[0021]

[0022] In the formula, G * The optimized complex domain generator, The optimized complex domain discriminator, λ is a hyperparameter, y and y′ are target echo data and target enhanced data respectively.

[0023] Preferably, the optimization process comprises the following steps:

[0024] S31, keep the generation capability of the complex domain generator unchanged, mark the real high SCR signal as 1, mark the false high SCR signal generated by the generator as 0, and train the complex domain discriminator;

[0025] S32, fix the network parameters of the complex domain discriminator, take the low SCR signal as input, and train the complex domain generator to obtain a high SCR signal;

[0026] S33, evaluating the generated high SCR signal using a complex domain discriminator, and optimizing the complex domain generator;

[0027] S34. Repeat S31 to S33 until a Nash equilibrium state is reached and a global optimal solution is obtained.

[0028] Preferably, in step S34, when the real signal space D s and the pseudo signal space D s The intersection of ' tends to balance in the game and satisfies D s ∩D' s =D' s =D s When p r =p ir =0.5;

[0029] in,

[0030]

[0031] In the formula, p r 、p ir They represent the correct discrimination probability and the incorrect discrimination probability of the complex domain discriminator respectively.

[0032] Preferably, in step S4, the angle is measured using a classical positioning estimation algorithm, which is specifically expressed as follows:

[0033]

[0034] In the formula, is the angle measurement result of DBF algorithm, R1=y'y 'H Represents the data covariance matrix of the target augmented data y'.

[0035] Therefore, the present invention adopts the above-mentioned high-precision angle measurement method of low-elevation-angle multipath removal of meter-wave radar based on feature game, which has the following technical effects:

[0036] (1) Using the generative adversarial network model, the direct echo signal contained in the target echo signal is extracted to separate the target echo signal from the multipath signal, thereby achieving the effect of enhancing the target echo signal;

[0037] (2) Based on feature game and multipath suppression, the complex domain generator and complex domain discriminator in the generative adversarial network model conduct game confrontation and optimization learning, so that the generated signal continuously approaches the real clutter signal and ensures that the energy of the output signal is not lost, thereby achieving high-precision positioning of low-altitude targets.

[0038] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 is a schematic diagram of receiving signals by a receiving array;

[0040] Figure 2 This is a framework diagram of the method in which the generator and the discriminator compete with each other;

[0041] Figure 3 It is a schematic diagram of the process of mutual optimization learning between the generator and the discriminator;

[0042] Figure 4 is the optimization direction of the complex field generator G;

[0043] Figure 5 is the optimization direction of the complex domain discriminator D;

[0044] Figure 6 It is a schematic diagram of the correct and incorrect probability of the discriminator D;

[0045] Figure 7 is the distribution diagram of the angle measurement error before and after unsupervised multipath suppression in scenario 1, where Figure 7 (a) is the distribution diagram of the angle error before multipath suppression. Figure 7 (b) is the distribution diagram of angle measurement error after multipath suppression;

[0046] Figure 8 This is a comparison chart of data characteristics before and after multipath suppression when the SNR is 0dB in scenario 1. Figure 8 (a) is the phase spectrum of sample 1, Figure 8 (b) is the spatial spectrum of sample 1, Figure 8 (c) is the phase spectrum of sample 2. Figure 8 (d) is the spatial spectrum of sample 2;

[0047] Fig. 9 This is a statistical analysis of the algorithm performance before and after multipath suppression in scenario 1;

[0048] Fig.10 is the distribution diagram of the angle measurement error before and after unsupervised multipath suppression in scenario 2, where Fig.10 (a) is the distribution diagram of the angle error before multipath suppression. Fig.10 (b) is the distribution diagram of angle measurement error after multipath suppression;

[0049] Fig.11 This is a comparison of data characteristics before and after multipath suppression when the SNR is -3dB in scenario 2, where: Fig.11 (a) is the phase spectrum of sample 1, Fig.11 (b) is the spatial spectrum of sample 1, Fig.11 (c) is the phase spectrum of sample 2. Fig.11 (d) is the spatial spectrum of sample 2;

[0050] Fig.12 This is a statistical analysis chart of the algorithm performance before and after multipath suppression in scenario 2. DETAILED DESCRIPTION

[0051] The present invention can be explained in more detail by the following examples. The purpose of disclosing the present invention is to protect all changes and improvements within the scope of the present invention. The present invention is not limited to the following examples.

[0052] The method for high-precision angle measurement of a meter-wave radar with low elevation angle and multipath removal based on feature game provided by the present invention comprises the following steps:

[0053] S1. Establish a signal model, as follows:

[0054] See also Figure 1 , considering that under actual position conditions, multipath echoes are scattered by irregular ground reflections, resulting in complex amplitude and phase distortions, which are not applicable to the traditional mirror reflection model. Therefore, according to the actual position reflection conditions, the received signal of the antenna array is expressed as:

[0055] y(t)=(a(θ d )+ρexp(-j2πΔR / λ)a(θ r ))s(t)+n(t),t=1,2,...,T

[0056] Where y(t) is the t-th snapshot data, a(θ d ) is the direct waveguide vector, a(θ r ) is the multipath steering vector, ρ is the complex reflection coefficient, s(t) is the unit signal complex envelope, and n(t) is the noise.

[0057] Simulate the direct waveguide vector a(θ d ), multipath steering vector a(θ r ) and complex reflection coefficient ρ, and then use the Gaussian random complex signal as the complex envelope of the signal to simulate the antenna receiving signal echo. In the model experiment, the complex terrain is set as a slope with a fixed angle to simulate the signal.

[0058] S2, see Figure 2, construct a generative adversarial network model to enhance the features of the target echo, as follows:

[0059] The generative adversarial network model includes a complex domain generator and a complex domain discriminator: the complex domain generator is used to extract the direct wave signal from the low signal-to-noise ratio echo signal and separate it from the multipath signal; the complex domain discriminator is used to mine the potential high-dimensional nonlinear features in the real high signal-to-noise ratio signal and distinguish the high signal-to-noise ratio signal generated by the complex domain generator from the real high signal-to-noise ratio signal by analyzing, comparing and synthesizing the high-dimensional nonlinear features. In the training stage, the radar echo data is input into the complex domain generator G of the generative adversarial network, and the complex domain discriminator D judges the waveform of the complex domain generator G. If it meets the direct echo signal, it is judged as 1, and if it does not meet the requirements, it is judged as 0. Through the game and optimization of the complex domain discriminator D and the complex domain generator G, the low SCR echo signal gradually tends to the high SCR signal. When the complex domain discriminator D cannot judge the generated high SCR signal and the real high SCR signal, the complex domain discriminator D constrains the complex domain generator G to enhance the characteristics of the signal and remove the undesired signal. In order to ensure that the generated signal satisfies the signal model and that the signal energy is not lost, it is also necessary to constrain the system power and signal distribution.

[0060] S3, see Figure 3 , through the game and optimization of the generator and the discriminator, the generated signal is close to the real high signal-to-noise ratio signal, reaching the Nash equilibrium state, realizing multipath suppression and signal feature enhancement, as follows:

[0061] (1) Training the complex domain discriminator: During the training process, the generation capability of the complex domain generator is kept unchanged. The complex domain discriminator marks the real high SCR signal as 1 and marks the false high SCR signal generated by the complex domain generator as 0. Through unsupervised learning, the signal discrimination capability of the complex domain discriminator is unilaterally increased.

[0062] Optimization objective function of complex domain discriminator D for:

[0063]

[0064] (2) Training the complex domain generator: Fix the network parameters of the complex domain discriminator of the t-th snapshot data, take the low SCR signal as input, and the complex domain generator outputs a high SCR signal that is close to the real high SCR signal waveform through learning.

[0065] The objective function of the complex domain generator G is to make the complex domain discriminator generate samples s for the complex domain generator t ''s discrimination probability is maximized, making log(1-D(s' c )) value is the smallest, and its objective function G * It can be expressed as:

[0066]

[0067] (3) Feature Game: The complex domain discriminator evaluates each generated signal and feeds the result back to the complex domain generator. The complex domain generator improves the output based on the result and continuously optimizes it to make the output effect more and more realistic.

[0068] During the optimization process, the complex domain generator continuously corrects the detailed characteristics of its output signal, making the output result of the complex domain generator more and more similar to the details of the real signal. The optimization direction is to make the pseudo signal space and the real signal space closer and closer. Please refer to Figure 4 Therefore, in the optimization process of the complex domain generator, D s ∩D' s It means that the complex domain generator extracts similar detailed features of the real signal and the input signal, and the range of the intersection will gradually increase with the optimization process.

[0069] The purpose of the complex domain discriminator is to distinguish between false high SCR signals and real high SCR signals. Its optimization direction is to separate the false signal space from the real signal space. Figure 5 Therefore, as the optimization process of the complex domain decision maker D s ∩D' s The range will gradually become smaller.

[0070] See also Figure 6 , the correct discrimination probability p of the complex domain discriminator r and the probability of wrong judgment p ir It is related to the overlapping area between the real signal space and the pseudo signal space. Since the overlapping area represents the similar features of the real signal space and the pseudo signal space, the complex domain discriminator cannot correctly judge the correctness of the overlapping area. Therefore, the correct judgment probability p of the complex domain discriminator is r and the probability of wrong judgment p ir Respectively expressed as:

[0071]

[0072] When the real signal space D s and the pseudo signal space D s The intersection of ' tends to balance in the game and satisfies D s ∩D' s =D' s =D s When , the discriminant probability relationship at this time is:

[0073] p r =p ir =0.5

[0074] At this time, the complex domain discriminator cannot accurately distinguish the high SCR signal generated by the complex domain generator and the real high SCR signal, and the discrimination probability is 50%, reaching the Nash equilibrium state; the complex domain generator can extract the effective real echo signal from the input low SCR radar echo signal to obtain a high SCR signal, and the generated signal distribution characteristics are consistent with the real signal characteristics, achieving feature enhancement and removing multipath, thereby improving the SCR.

[0075] (4) Iterative optimization: Repeat the above steps, alternately iteratively optimize the complex domain discriminator and the complex domain generator, so that the two can reach the optimal solution in the game.

[0076] The objective function of unsupervised multipath suppression is:

[0077]

[0078] When the complex domain generator G is fixed, the complex domain discriminator When, m D axV(D,G) achieves the maximum value; among them, are the probability density of the real high SCR signal sample and the probability density of the pseudo high SCR signal sample generated by the complex domain generator. Substituting the maximum value of the complex domain discriminator into the objective function of unsupervised multipath suppression, we can obtain The minimum value of

[0079]

[0080] Solving the equation, we can get the objective function of the complex domain generator G is to minimize the JS divergence between the generated sample set and the real sample set, and the minimum value is -log4. Complex Domain Discriminator Approaching the stable point, the training ends and reaches the global optimal solution.

[0081] In addition, in order to ensure that the generated signal conforms to the objective physical scattering characteristics, the distribution characteristics and power of the generated signal are constrained to ensure that the generated signal meets the signal model and there is no energy loss, as follows:

[0082] In order to enable the neural network to effectively and accurately mine the real target echo signal contained in the input signal y, the optimization function of the neural network must constrain the characteristics and power of the generated signal to ensure that the signal generated by the complex domain generator is close to the real clutter signal input by the complex domain discriminator, and the energy of the output signal is not lost, that is:

[0083]

[0084] Among them, the high SCR signal y' generated by the complex domain generator is consistent with the distribution characteristics of the real high SCR signal, and the complex domain discriminator D cannot accurately distinguish it, and then enhances the characteristics of the target echo signal.

[0085] S4, after training the complex domain generator G and complex domain discriminator D models with big data, the target echo signal can be effectively mined from the low SCR echo signal, thereby realizing the separation of the target echo signal and the multipath signal, and thus improving the SCR of the output signal. Therefore, the signal after multipath suppression and feature enhancement is input into the classic positioning estimation algorithm (such as DBF, MUSIC, etc.) to obtain the angle measurement of the meter-wave radar and realize the high-precision positioning of low-altitude targets, as follows:

[0086]

[0087] In the formula, is the angle measurement result of DBF algorithm, R1=y'y 'H Represents the data covariance matrix of the target enhanced data y'. Since the multipath signal is greatly suppressed, the target features are effectively enhanced, and the accuracy of the positioning estimation algorithm in the multipath environment is significantly improved.

[0088] Simulation and experimental analysis

[0089] In the simulation experiment, two groups of experiments are set up. One is the performance analysis of the algorithm under the condition of matching signal-to-noise ratio, and the other is the generalization analysis of the algorithm to the signal-to-noise ratio under the condition that the signal-to-noise ratio of the training set and the test set is not completely matched. The details are as follows:

[0090] Scenario 1: Impact of signal-to-noise ratio matching on angle measurement accuracy

[0091] The array structure in this scenario is an equidistant array with M = 24 elements, wavelength λ = 1 meter, array spacing d = λ / 2, and the direct signal and multipath signal are coherent sources. Considering that the 3dB beam width is about 4°, the direct wave incident angle θ d ~U(1°,3°), the multipath reflected wave incident angle satisfies θ r =θ d -0.4°, after pulse compression and multi-pulse accumulation, the signal-to-noise ratio of the training set and test set data array element level is 0dB~10dB. During the experiment, 100 Monte Carlo experiments were performed on the training set and the test set, respectively, with 6000 samples for training and testing, and two angle measurement algorithms (DBF and MUSIC) were used for angle measurement.

[0092] Meter-wave radar low elevation angle multipath removal high-precision angle measurement technology is applied to the distribution of angle measurement errors before and after unsupervised multipath suppression. Figure 7, where the horizontal axis represents the sample label and the vertical axis represents the error angle. It can be seen that by adopting the feature game mechanism, the angle measurement error can be greatly reduced, and the angle measurement error is mainly concentrated in the range of ±1°.

[0093] Meter-wave radar low elevation angle multipath removal high-precision angle measurement technology is applied when the SNR is 0dB. For the results of the data feature comparison before and after multipath suppression, please refer to Figure 8 , where the horizontal coordinates are the antenna index and the direction estimation angle, and the corresponding vertical coordinates are the normalized phase and the normalized spectrum. It can be seen that by adopting the method provided by this embodiment, the target signal can be more effectively mined, and the signal characteristics meet the far-field model, and the real elevation angle of the target can be measured more accurately.

[0094] The results of the statistical analysis of the algorithm performance before and after the multipath suppression are shown in the application of the high-precision angle measurement technology of low-elevation-angle multipath suppression by meter-wave radar. Fig. 9 , where the abscissa is the signal-to-noise ratio and the ordinate is the root mean square error. It can be seen that the method of this embodiment can significantly improve the performance of the angle measurement algorithm, and the performance of the DBF algorithm after multipath suppression is better than that of the SSMUSIC algorithm.

[0095] Scenario 2: Impact of signal-to-noise ratio mismatch on angle measurement accuracy

[0096] The array structure in this scenario is an equidistant array with M = 24 elements, wavelength λ = 1 meter, array spacing d = λ / 2, and the direct signal and multipath signal are coherent sources. Considering that the 3dB beam width is about 4°, the direct wave incident angle θ d ~U(1°,3°), the multipath reflected wave incident angle satisfies θ r =θ d -0.4°, after pulse compression and multi-pulse accumulation, the signal-to-noise ratio of the training set data array element level is 0dB~10dB, and the signal-to-noise ratio of the test set data array element level is -3dB~7dB. During the experiment, 100 Monte Carlo experiments were performed on the training set and the test set, respectively, with 6000 samples for training and testing, and two angle measurement algorithms (DBF and MUSIC) were used for angle measurement.

[0097] The high-precision angle measurement technology of low-elevation-angle multipath removal provided in this example is applied to the distribution of angle measurement errors before and after unsupervised multipath suppression. Fig.10 , where the horizontal axis represents the sample label and the vertical axis represents the error angle. It can be seen that by adopting the feature game mechanism, the angle measurement error can still be greatly reduced under the condition of mismatched signal-to-noise ratio, and the angle measurement error is mainly concentrated in the range of ±1.5°. Compared with the error distribution under matched signal-to-noise ratio, on the one hand, the signal-to-noise ratio is lower and the error is larger under mismatched conditions, and on the other hand, parameter mismatch will also have a certain performance impact.

[0098] The high-precision angle measurement technology for low-elevation multipath suppression provided in this example is applied when the SNR is -3dB. For the comparison of data characteristics before and after multipath suppression, please refer to Fig.11 , where the horizontal coordinates are the antenna index and the direction estimation angle, and the corresponding vertical coordinates are the normalized phase and the normalized spectrum. It can be seen that by adopting the method of this embodiment, the target signal can be more effectively mined in the case of signal-to-noise ratio mismatch, and the signal characteristics meet the far-field model, so that the real elevation angle of the target can be measured more accurately.

[0099] The results of the statistical analysis of the algorithm performance before and after the multipath suppression are shown in the application of the high-precision angle measurement technology of low-elevation-angle multipath suppression provided in this example. Fig.12 , where the abscissa is the signal-to-noise ratio and the ordinate is the root mean square error. It can be seen that by adopting the method of this embodiment, the performance of the angle measurement algorithm can still be significantly improved in the case of signal-to-noise ratio mismatch, and the performance of the DBF algorithm after multipath suppression is better than that of the SSMUSIC algorithm. Compared with the traditional estimation method, the method of this embodiment can still maintain a high accuracy in the case of signal-to-noise ratio mismatch, indicating that it has strong generalization and practicality in practical applications.

[0100] Therefore, the present invention adopts the above-mentioned low-elevation-angle multipath removal and high-precision angle measurement method of meter-wave radar based on feature game, which can improve the angle measurement accuracy of meter-wave radar when the signal-to-noise ratio is matched or incompletely matched, and provides new possibilities for the application of feature game mechanism in this field.

[0101] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A high-precision angle measurement method for low-elevation-angle multipath removal of meter-wave radar based on feature game, characterized in that: The following steps are involved: S1. Establish a receiving signal model of the antenna array according to the actual position reflection conditions; S2, build a generative adversarial network model; S3. Based on feature game and multipath suppression, the generative adversarial network model is optimized to enhance the target echo signal; S4. Based on the enhanced signal, the angle is measured using the classical positioning estimation algorithm.

2. The method for high-precision angle measurement of low-elevation-angle multipath elimination of meter-wave radar based on feature game according to claim 1 is characterized in that: In step S1, the expression of the received signal model is as follows: y(t)=(a(θ d )+ρexp(-j2πΔR / λ)a(θ r ))s(t)+n(t),t=1,2,...,T Where t is the number of snapshots, y(t) is the tth snapshot data, and a(θ d ) is the direct waveguide vector, a(θ r ) is the multipath steering vector, ρ is the complex reflection coefficient, s(t) is the unit signal complex envelope, and n(t) is the noise.

3. The method for high-precision angle measurement of low-elevation-angle multipath elimination of meter-wave radar based on feature game according to claim 1 is characterized in that: In step S2, the generative adversarial network model includes a complex domain generator and a complex domain discriminator, as follows: The complex domain generator is used to extract the direct wave signal from the low signal-to-noise ratio echo signal and separate it from the multipath signal; the complex domain discriminator is used to explore the potential high-dimensional nonlinear features in the real high signal-to-noise ratio signal, and distinguish the high signal-to-noise ratio signal generated by the complex domain generator from the real high signal-to-noise ratio signal by analyzing, comparing and integrating the high-dimensional nonlinear features.

4. The method for high-precision angle measurement of low-elevation-angle multipath elimination of meter-wave radar based on feature game according to claim 3 is characterized in that: In step S3, an unsupervised multipath suppression method is used to optimize the complex domain discriminator and the complex domain generator, and the objective function is: In the formula, G is the complex domain generator, D is the complex domain discriminator, and s t 、s t ' are real signal samples and generated signal samples respectively, D s , D s ' are the real signal space and the pseudo signal space respectively.

5. The method for high-precision angle measurement of low-elevation-angle multipath elimination of meter-wave radar based on feature game according to claim 4 is characterized in that: The optimization of the generative adversarial network model involves constraining the characteristics and power of the generated signal as follows: In the formula, G * The optimized complex domain generator, The optimized complex domain discriminator, λ is a hyperparameter, y and y′ are target echo data and target enhanced data respectively.

6. The method for high-precision angle measurement of low-elevation-angle multipath elimination of meter-wave radar based on feature game according to claim 5 is characterized in that: The optimization process includes the following steps: S31, keeping the generation capability of the complex domain generator unchanged, marking the real high SCR signal as 1, marking the false high SCR signal generated by the complex domain generator as 0, and training the complex domain discriminator; S32, fix the network parameters of the complex domain discriminator, take the low SCR signal as input, and train the complex domain generator to obtain a high SCR signal; S33, evaluating the generated high SCR signal using a complex domain discriminator, and optimizing the complex domain generator; S34. Repeat S31 to S33 until a Nash equilibrium state is reached and a global optimal solution is obtained.

7. According to the method for high-precision angle measurement of low-elevation-angle multipath elimination of meter-wave radar based on feature game in claim 6, in step S34, when the real signal space D s and the pseudo signal space D s The intersection of ' tends to balance in the game and satisfies D s ∩D s '=D s '=D s When p r =p ir =0.5; in, In the formula, p r 、p ir They represent the correct discrimination probability and the incorrect discrimination probability of the complex domain discriminator respectively.

8. The method for high-precision angle measurement of low-elevation-angle multipath elimination of meter-wave radar based on feature game according to claim 7 is characterized in that: In step S4, the angle is measured using a classical positioning estimation algorithm, which is specifically expressed as follows: In the formula, is the angle measurement result of DBF algorithm, R1=y'y 'H Represents the data covariance matrix of the target augmented data y'.

Images