Unmanned aerial vehicle data link complex electromagnetic interference sensitive parameter extraction method and related device

Abstract

The invention provides a method for extracting complex electromagnetic interference sensitive parameters of a data link of an unmanned aerial vehicle and a related device, and the method comprises the steps: initializing an observation data set, observation data in the observation data set comprises an interference signal parameter group and a signal-to-interference ratio threshold which causes the unmanned aerial vehicle to lose lock of a data link under the electromagnetic interference of the interference signal parameter group; constructing a target function by taking the interference signal parameter group as an independent variable and taking signal-to-interference ratio threshold maximization as an optimization target, and constructing a proxy model of the target function by adopting a Gaussian process; inputting the interference signal parameter group in the current observation data set into the proxy model, and determining posterior characteristic parameters; constructing an acquisition function based on the posterior characteristic parameters, selecting an interference signal parameter group in the search space to carry out an electromagnetic interference experiment, determining a corresponding signal-to-interference ratio threshold value, and updating the current observation data set; the above process is repeated until the preset convergence condition is met, the electromagnetic sensitivity parameter combination can be obtained more quickly, and the manpower, material and time costs of the electromagnetic sensitivity test are reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned aerial vehicles, and in particular to a method for extracting sensitive parameters of complex electromagnetic interference of a data link of an unmanned aerial vehicle and related devices. BACKGROUND

[0002] Test design is the basis for determining the electromagnetic sensitivity of equipment. Affected by the highly complex electromagnetic environment of modern battlefields and the nonlinear response of frequency equipment, the electromagnetic sensitivity test of unmanned aerial vehicles faces multi-dimensional and complex challenges. Optimizing test design to determine the sensitive interference source parameters of the data link of an unmanned aerial vehicle under a specific test scenario is the basis for improving the complex electromagnetic environment adaptability of the data link and is conducive to achieving higher reliability of the unmanned aerial vehicle during the mission cycle. Therefore, it is necessary to develop a method for extracting electromagnetic interference sensitive parameters of the data link of an unmanned aerial vehicle under a complex electromagnetic environment.

[0003] Traditional electromagnetic sensitivity tests usually use a single frequency continuous wave to interfere with the test equipment. However, under a complex electromagnetic environment, the interference signal is usually not a single signal, and the method of using parameter scanning for sensitivity test is no longer applicable. SUMMARY

[0004] The embodiments of the present application provide a method for extracting sensitive parameters of complex electromagnetic interference of a data link of an unmanned aerial vehicle and related devices to solve the problem of inapplicable electromagnetic interference sensitive parameters obtained by using parameter scanning for sensitivity test.

[0005] In a first aspect, the embodiments of the present application provide a method for extracting sensitive parameters of complex electromagnetic interference of a data link of an unmanned aerial vehicle, comprising:

[0006] Step 1: initialize an observation data set, the observation data set comprising at least one observation data, the observation data comprising a group of interference signal parameters and a signal-to-interference ratio threshold value at which the data link of the unmanned aerial vehicle loses lock under electromagnetic interference of the group of interference signal parameters; the group of interference signal parameters comprising a plurality of electromagnetic interference signal parameters;

[0007] Step 2: construct a target function with the group of interference signal parameters as the independent variable and the maximization of the signal-to-interference ratio threshold value at which the data link loses lock as the optimization goal, and construct a proxy model of the target function using a Gaussian process;

[0008] Step 3: input the group of interference signal parameters in the current observation data set into the proxy model to determine the posterior characteristic parameters of the Gaussian process for the observed value of the signal-to-interference ratio threshold value of the group of interference signal parameters; the posterior characteristic parameters comprising a mean value and / or a variance;

[0009] Step 4: constructing an acquisition function based on the posterior characteristic parameter, and selecting an interference signal parameter group as a next test point in a search space by maximizing the acquisition function to perform an electromagnetic interference experiment, determining a signal-to-interference ratio threshold corresponding to the next test point, and storing the next test point and the corresponding signal-to-interference ratio threshold as a set of observation data into a current observation data set;

[0010] Step 5: returning to Step 3 to repeat the execution until a preset convergence condition is met, and outputting an interference signal parameter group corresponding to a maximum signal-to-interference ratio threshold when the data link is lost.

[0011] In a second aspect, an embodiment of the present application provides a device for extracting complex electromagnetic interference sensitive parameters of a data link of a UAV, comprising:

[0012] An observation data set initialization module is configured to initialize an observation data set, wherein the observation data set comprises at least one observation data, the observation data comprises an interference signal parameter group and a signal-to-interference ratio threshold when the data link of the UAV is lost under electromagnetic interference of the interference signal parameter group, and the interference signal parameter group comprises a plurality of electromagnetic interference signal parameters.

[0013] A proxy model construction module is configured to construct a target function by taking the interference signal parameter group as an independent variable and taking the maximum signal-to-interference ratio threshold when the data link is lost as an optimization target, and construct a proxy model of the target function by using a Gaussian process.

[0014] A posterior characteristic parameter acquisition module is configured to input the interference signal parameter group in the current observation data set into the proxy model, and determine a posterior characteristic parameter of an observed value of the signal-to-interference ratio threshold of the interference signal parameter group by the Gaussian process; the posterior characteristic parameter comprises a mean value and / or a variance.

[0015] An observation set updating module is configured to construct an acquisition function based on the posterior characteristic parameter, and select an interference signal parameter group as a next test point in a search space by maximizing the acquisition function to perform an electromagnetic interference experiment, determine a signal-to-interference ratio threshold corresponding to the next test point, and store the next test point and the corresponding signal-to-interference ratio threshold as a set of observation data into a current observation data set.

[0016] An interference signal parameter group output module is configured to repeat the above steps until a preset convergence condition is met, and output an interference signal parameter group corresponding to a maximum signal-to-interference ratio threshold when the data link is lost.

[0017] In a third aspect, an embodiment of the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, and the processor implements the steps of the method in any possible implementation manner of the first aspect.

[0018] In a fourth aspect, the embodiments of the present application provide a computer readable storage medium, which stores a computer program, and the computer program, when executed by a processor, implements the steps of the method in any possible implementation manner of the first aspect.

[0019] The unmanned aerial vehicle data link complex electromagnetic interference sensitive parameter extraction method provided by the embodiments of the present application applies the Bayesian optimization method to constructing the unmanned aerial vehicle data link learning type electromagnetic sensitivity test, is a high-efficiency data-driven test method, can more quickly obtain an electromagnetic sensitive parameter combination, reduces the manpower, material resources and time cost of the electromagnetic sensitivity test, and lays a foundation for the effect research of the data link under complex electromagnetic interference signals. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 is a working scene diagram of an unmanned aerial vehicle data link provided by the embodiments of the present application;

[0021] Figure 2 is a configuration schematic diagram of an electromagnetic interference injection test method provided by the embodiments of the present application;

[0022] Figure 3 is an implementation flowchart of an unmanned aerial vehicle data link complex electromagnetic interference sensitive parameter extraction method provided by the embodiments of the present application;

[0023] Figure 4 is a signal-to-interference ratio convergence diagram of a data link under microwave pulse loss provided by the embodiments of the present application;

[0024] Figure 5 is a frequency spectrum schematic diagram of a high-repetition-frequency microwave pulse provided by the embodiments of the present application;

[0025] Figure 6 is a waveform schematic diagram of an intermediate frequency signal of a data link under non-interference signal conditions and microwave pulse interference conditions provided by the embodiments of the present application;

[0026] Figure 7 is a principle schematic diagram of a data link receiver data terminal provided by the embodiments of the present application;

[0027] Figure 8 is a double-source electromagnetic interference sensitivity test flowchart provided by the embodiments of the present application;

[0028] Figure 9 is a convergence diagram of a sum of signal-to-interference ratios of a data link under double-source interference provided by the embodiments of the present application;

[0029] Figure 10 is a relationship curve schematic diagram of a loss signal-to-interference ratio threshold value and an interference signal center frequency provided by the embodiments of the present application;

[0030] Figure 11is a relationship curve between the data link loss of lock signal to interference ratio and the interference bandwidth under the wideband white noise interference provided by the embodiment of the present application. DETAILED DESCRIPTION

[0031] The embodiment of the present application will be described in detail below with reference to the drawings.

[0032] Figure 1 The unmanned aerial vehicle data link working scene schematic diagram provided by the embodiment of the present application is shown in FIG. 1. Figure 1 As shown in FIG. 1, the unmanned aerial vehicle communicates with the ground control station through the data link, and the ground control station transmits remote control signals to control the flight path and assign tasks to the unmanned aerial vehicle. During the execution of the task, the unmanned aerial vehicle may be subjected to electromagnetic interference from multiple directions on the ground or in the air. In general, external electromagnetic radiation interference may affect the frequency equipment through the "front door" or "back door" coupling path. The data link equipment provided by the embodiment of the present application is mounted on a medium-sized unmanned aerial vehicle, and the internal shielding measures are relatively complete, so the front door coupling is the main interference coupling mode. When the uplink remote control link signal is interfered, the unmanned aerial vehicle is prone to lose control and combat capability, so it is necessary to focus on the sensitivity test of the interference signal near the uplink working frequency band of the data link.

[0033] Specifically, the unmanned aerial vehicle data link assembled in the embodiment is mounted on a medium-sized unmanned aerial vehicle, which adopts a frequency division multiplexing working mode in the L band. The data link sets multiple alternative working channels in a specific frequency band, and the research of the embodiment is carried out on the main working channel with a center frequency f0. The unmanned aerial vehicle data link equipment mainly consists of an airborne antenna, a duplexer, a transmitting unit, a receiving unit and a data terminal, and the basic structure is shown in FIG. 2. Figure 1

[0034] During the flight of the unmanned aerial vehicle, the airborne antenna receives a working signal with a power Pr and an interference signal with a power Pj. The ratio of the sum of the powers of the received signal, the interference and the noise signal is

[0035]

[0036] In formula (1), N is the noise power of the receiver, N=kT0B n F ns , k is the Boltzmann constant, T0=290K is the standard thermal noise temperature, B n is the noise bandwidth of the receiver, kT0B n represents the thermal noise of the receiver, F ns is the noise factor of the receiving system. Denote the output signal power density of the receiving unit as C and the noise density as N0. If the data link normally communicates, the following needs to be met:

[0037]

[0038] ​In formula (2), R b is the communication data rate, is the minimum ratio of bit energy to noise density to meet the specified bit error rate of the demodulator. The spread spectrum processing gain of the data link receiver is Assuming that the received interference signal power is much greater than the noise power, i.e. P j > kT0B n F ns , the signal-to-interference ratio P SIR = P r / P j , (E b / N0) rec is the actual ratio of bit energy to noise density. Then according to formula (2), the condition for the data link to communicate normally is

[0039]

[0040] Therefore, when the signal-to-interference ratio P SIR of the UAV data link decreases to a certain limit, the actual ratio of bit energy to noise density (E b / N0) rec decreases, and the data link communication quality decreases. When (E b / N0) rec decreases to less than the minimum ratio of bit energy to noise density to meet the specified bit error rate of the demodulator, the data link communication faces the risk of interruption. When P SIR continues to decrease until the data frame cannot achieve synchronization, the data link will lose lock. Loss of lock is a relatively serious communication accident, and often means that the communication between the UAV and the ground station is completely interrupted. The electromagnetic sensitivity test in the present embodiment takes the signal-to-interference ratio when the data link loses lock as the criterion for the sensitivity of the data link to electromagnetic interference.

[0041] Since the UAV data link is difficult to test the electromagnetic sensitivity under flight conditions, the injection method test can quickly and accurately, repeatedly find the electromagnetic signal that makes it sensitive, and can ensure good consistency between the injection and radiation test results. The configuration of the electromagnetic interference injection test method is as follows: Figure 2As shown, the airborne and ground station data link is powered by a DC regulated power supply. In order to simulate the channel state of long distance communication in the air, two fixed attenuators and an adjustable attenuator are connected in series between the airborne and ground station data link through a radio frequency cable. The fixed attenuators protect the radio frequency front end from being damaged by sudden high power interference, and the spatial attenuation of the communication signal under different flight distances of the unmanned aerial vehicle is simulated by adjusting the adjustable attenuator. The signal generator injects interference signals into the data link communication link through a combiner for electromagnetic interference on the uplink. The number of signal generators can be configured according to the number of electromagnetic interference sources simulated in the test. The interference signal power arriving at the airborne data link antenna port is monitored by a spectrum analyzer at the data link antenna port, and the ground station data link is connected to the monitoring software on the monitoring computer through a network cable. Whether the data link is locked can be observed through the monitoring software.

[0042] The traditional electromagnetic susceptibility test usually uses a single frequency continuous wave to interfere with the test equipment. Since the continuous wave has only two variable parameters of frequency and power, the test process is divided into two steps: (1) set the signal generator to sweep mode, preset the frequency variation interval, and perform frequency scanning until the data link loses lock effect to obtain the sensitive interference frequency of the data link; (2) at the measured sensitive frequency, reduce the interference power to the normal working of the data link, and then gradually increase the signal source output power until the data link loses lock again to obtain the interference threshold of the data link at the frequency. The single frequency continuous wave interference can quickly scan the sensitive frequency and power threshold, but in a complex electromagnetic environment, the interference signal may have different types of combinations, and there are multiple variable parameters. The test number required to find the most sensitive interference parameter combination of the data link will increase exponentially. The parameter scanning method is no longer suitable for sensitivity test. Therefore, the embodiment provides a method for extracting sensitive parameters of complex electromagnetic interference of unmanned aerial vehicle data link, thereby improving the test efficiency of multi-parameter electromagnetic interference.

[0043] Referring to Figure 3 which shows the implementation flowchart of the method for extracting sensitive parameters of complex electromagnetic interference of unmanned aerial vehicle data link provided by the embodiment of the application, and is described in detail as follows:

[0044] S101: initialize an observation data set, the observation data set comprising at least one observation data, the observation data comprising an interference signal parameter group and a signal-to-interference ratio threshold at which the data link loses lock under electromagnetic interference of the interference signal parameter group; and the interference signal parameter group comprising a plurality of electromagnetic interference signal parameters.

[0045] In the embodiment, the target of the sensitivity test under complex electromagnetic interference is to search for the parameter combination of the multi-dimensional electromagnetic interference signal that makes the data link most sensitive. It can be obtained from equation (3) that whether the data link is electromagnetically sensitive can be measured by the signal-to-interference ratio of the antenna reception. Therefore, under a specific working signal, the interference signal parameter combination that makes the data link reach the maximum signal-to-interference ratio of loss of lock is the interference combination that makes the data link most sensitive.

[0046] In one possible implementation, the specific implementation process of initializing the observation data set in S101 includes:

[0047] a plurality of interference signal parameter groups are randomly generated in the parameter search space, and an electromagnetic interference test of the unmanned aerial vehicle is performed on each interference signal parameter group by using the electromagnetic interference injection test method to determine the signal-to-interference ratio threshold of the loss of lock of the data link of the unmanned aerial vehicle under the electromagnetic interference of each interference signal parameter group;

[0048] Based on the formula y i =R(x i )+ε i ,i=1,…,N, the observation value corresponding to each interference signal parameter group is determined, and the interference signal parameter group and the corresponding observation value are combined to form an observation data, and an observation data set is generated;

[0049] wherein R(x i ) represents the signal-to-interference ratio threshold, ε i represents additive noise with a distribution of x i represents the i th interference signal parameter group in the observation data set, and y i represents the observation value of the i th interference signal parameter group.

[0050] Specifically, assuming that the initial parameter set of the electromagnetic interference signal is X={x1,x2,…x N}, the observation data set of the interference signal parameter and the electromagnetic sensitivity threshold is D={(x i ,y i ),i=1,…,N}, wherein y i is the observation value; y i =R(x i )+ε i ,i=1,…,N; and in the formula, ε i represents additive noise with a distribution of The input data X={x1,x2,…x N} and the observation data y={y1,y2,…y N} are combined to form the observation data set D={(x i ,y i ),i=1,…,N}.

[0051] S102: Construct a target function with the interference signal parameter set as the independent variable and the threshold value of the signal-to-interference ratio at which the data link is lost maximized as the optimization objective, and construct a surrogate model of the target function using a Gaussian process.

[0052] The essence of the electromagnetic susceptibility test design is a single-objective optimization problem, which aims to obtain an interference signal parameter combination x that maximizes the global threshold value of the signal-to-interference ratio at which the data link is lost. * Therefore, the target function of the present embodiment is as follows:

[0053]

[0054] In formula (4), the vector x represents the parameters of the electromagnetic interference source. In general, each interference source parameter x is limited within a specific search space χ according to the parameter range concerned in the actual test. R(x) is the signal-to-interference ratio threshold at which the data link is lost under electromagnetic interference.

[0055] The optimization algorithm is an iterative process for solving black-box global optimization problems. Such optimization problems have high data collection costs, and therefore can only be tested and evaluated a small number of times. In order to make full use of the scarce data, the Bayesian optimization method first selects a surrogate model to assume the posterior distribution of the target function.

[0056] In one possible implementation, the specific process of constructing the surrogate model in step S102 includes:

[0057] The mean function and the kernel function of the Gaussian process are set as the prior distribution; wherein the Gaussian process is defined as:

[0058] R(x) ~ GP(m(x), k θ (x i ,x j ))(5)

[0059] In formula (5), R(x) represents the regression function, m(x) represents the mean function, and is set to zero; k θ (x i ,x j ) represents the kernel function, and a represents the output scaling factor, x i and x j represent the i-th interference signal parameter set and the j-th interference signal parameter set in the observation data set, respectively; r = ||x i -x j ||, and l represents the characteristic length scale;

[0060] For the input value x of the Gaussian process, the regression function and the observation value follow a joint Gaussian prior distribution:

[0061]

[0062] In formula (6), k θ (x, X) = [k θ (x, x1), k θ (x, x2), …, k θ (x, x N )];

[0063] The conditional posterior distribution of R(x) is: where μ(x) represents the mean of the Gaussian process, σ 2 (x) represents the variance of the Gaussian process; and:

[0064]

[0065] In formula (7), X represents a set of interference signal parameter groups; K represents an N×N covariance matrix between input values, D represents an observation data set, y represents an observation value, I represents a unit matrix; k θ (x, X) represents a covariance vector between a newly input interference signal parameter group and the set of interference signal parameter groups; k θ (x, x N ) represents the covariance between the newly input interference signal parameter group and the Nth interference signal parameter group in the set of interference signal parameter groups.

[0066] Specifically, in the case of using a Gaussian process as a surrogate model, f(x) conforms to a Gaussian prior distribution specified by its mean function m(x) and kernel function k θ (x, x′), that is, Since there is no available prior information, m(x) is set to zero at this time. The vector y = {R(x i )+ε i , i = 1, …, N follows a multivariate Gaussian distribution:

[0067]

[0068] In formula (8), K is an N×N covariance matrix between training points (interference signal parameter groups), that is, K θ＝kθ (x i , x j ), 1≤ij≤N, and I is a unit matrix.

[0069] The commonly used kernel function is the Matern function with a smoothing factor of 2.5, and its value is:

[0070]

[0071] In formula (9), a is an output scaling factor, r = ‖x i -x j ‖ is the distance between points xi distance to x j , l is the characteristic length scale.

[0072] For an input set of interference signal parameters x, the electromagnetic sensitivity threshold R(x) and the observed response y follow a joint Gaussian prior distribution:

[0073]

[0074] In equation (10), k θ (x, x1), k θ (x, x2), …, k θ (x, x θ (x, x N ). The conditional posterior distribution of R(x) within the interval D of the trial observations can be expressed as:

[0075]

[0076] In equation (11), the mean μ and the variance σ 2 of the posterior distribution of the Gaussian process at x are calculated as:

[0077]

[0078] In one possible implementation, the method provided by the embodiment further includes:

[0079] estimating the hyperparameters of the surrogate model by maximizing the log marginal likelihood function;

[0080] optimizing the hyperparameters of the surrogate model by iteratively optimizing the log marginal likelihood function using the gradient descent method.

[0081] In one possible implementation, the expression of the log marginal likelihood function is:

[0082]

[0083] In equation (13), θ represents the set of hyperparameters, and X represents the set of interference signal parameter sets, σ2 represents the noise variance, l represents the characteristic length scale, a represents the output scaling factor, and N represents the number of observed data in the observed data set; K represents the N x N covariance matrix between the input values, and I represents the unit matrix.

[0084] In the embodiment, the surrogate model of the Gaussian process includes some hyperparameters, the noise variance is used to capture the change of the measurement threshold, the output scaling factor a is used to calibrate the change, and the length scale l is used to capture the smoothness or variability of the objective function. For the sake of simplicity, the vector To represent the set of hyperparameters learned during model training, we estimate the hyperparameters by maximizing the log-marginal likelihood function and optimize it using gradient descent.

[0085] S103: Input the interference signal parameter group in the current observation data set into the proxy model to determine the posterior characteristic parameters of the observation value of the signal-to-interference ratio threshold of the interference signal parameter group by the Gaussian process; the posterior characteristic parameters include mean and / or variance.

[0086] S104: Constructing an acquisition function based on the posterior characteristic parameters, and selecting an interference signal parameter group as the next test point in the search space by maximizing the acquisition function to perform the electromagnetic interference experiment, determining a signal-to-interference ratio threshold corresponding to the next test point, and storing the next test point and the corresponding signal-to-interference ratio threshold as a set of observation data in the current observation data set.

[0087] In one possible implementation, the acquisition function includes an improved probability acquisition function:

[0088] The expression of the improved probability acquisition function is:

[0089] Among them, f(x) represents the mapping relationship between the interference signal parameter group and the signal-to-interference ratio threshold, x + represents the current optimal value of the objective function, μ(x) represents the mean of the Gaussian process, σ 2 (x) represents the variance of the Gaussian process; Φ(·) represents the cumulative distribution function, and D represents the observed data set.

[0090] In one possible implementation, the acquisition function includes a desired improved acquisition function;

[0091] The expression of the expected improved acquisition function is: in, φ(z) represents the probability density function; f(x) represents the mapping relationship between the interference signal parameter group and the signal-to-interference ratio threshold, x + represents the current optimal solution of the objective function, μ(x) represents the mean of the Gaussian process, and σ(x) represents the standard deviation of the Gaussian process.

[0092] In this embodiment, after obtaining the posterior distribution of the threshold of the out-of-lock signal-to-interference ratio R(x), the posterior distribution is converted into an optimized objective function by using an acquisition function u(·). The acquisition function is used to guide the search for the maximum out-of-lock signal-to-interference ratio to the potential high-value point of the objective function, so as to serve as a reference for the next sampling point. In the process of selecting the next sampling point by using the acquisition function, the relationship between exploitation and exploration needs to be balanced. Exploitation refers to sampling in the vicinity of the solution represented by the current surrogate model, so as to make full use of the existing information. Exploration refers to sampling in the less explored area of the search space, in which the prediction uncertainty of the base surrogate model is high, and there may be a global maximum value. If the selected acquisition function pays too much attention to exploitation, the prediction result will be limited to a local maximum value, and the global search will be ignored. If the acquisition function pays too much attention to exploration, it will tend to extensive search, and cannot concentrate on the maximum value. An ideal acquisition function can balance exploitation and sampling, and fully search the parameter space without missing the global maximum value.

[0093] The selected acquisition function in this embodiment can include probability of improvement (PI), expected improvement (EI), upper confidence bound (UCB), lower confidence bound (LCB), and the like. The UCB and LCB are respectively used for searching the maximum and minimum target values.

[0094] wherein PI tends to exploitation, and does not consider the potential improvement when assigning a value to a new sampling point. The expression of the PI acquisition function is as follows:

[0095]

[0096] In formula (14), x + is the optimal value of the current objective function, μ(x) and σ(x) are the mean and standard deviation of the Gaussian process, and Φ(·) is the cumulative distribution function.

[0097] Compared with the PI function, the EI measures the expectation of the prediction distribution on the improvement of the objective function, and the expression is as follows:

[0098]

[0099] In formula (15), φ(·) is the probability density function.

[0100] UCB function is used for search of the maximization problem, which only depends on the posterior distribution of the Gaussian process, and does not depend on the prior observation value, so it is more suitable for dynamic changing scene. The expression of UCB is:

[0101] u UCB (x|D)=μ(x)+κσ(x) (16)

[0102] In formula (16), κ is a weight balance between utilization and exploration, κ≥0. The higher the value of κ, the more emphasis on exploration by expanding the uncertainty of the model. That is, the next set of sampling points x is updated by maximizing the acquisition function N+1 : x N+1 =argmaxu(X|D).

[0103] S105: return to step S103 for repeated execution until a preset convergence condition is met, and output the interference signal parameter set corresponding to the maximum data link loss lock signal to interference ratio threshold.

[0104] In one possible embodiment, the specific implementation process of S101 further includes:

[0105] The input observation data is standardized to zero mean and unit variance to improve the convergence speed and prediction accuracy of the experimental design modeling.

[0106] As can be seen from the above embodiment, the embodiment provides a method for extracting sensitive parameters of complex electromagnetic interference of unmanned aerial vehicle data link. First, the initial observation data set is obtained by random generation, and the input observation data is standardized to zero mean and unit variance to improve the convergence speed and prediction accuracy of the experimental design modeling. These input data are used to train the surrogate model, and the surrogate model is constructed by combining the prior observation and the prior function, and the Gaussian process is used as the surrogate model. The prior function captures the hypothesis of the response surface (such as smoothness and test noise) to infer the global optimal position. After the training of the surrogate model is completed, the new test points are selected in the search space by optimizing the acquisition function. Next, the interference signal parameter set proposed by the acquisition function is tested, and the results are appended to the observation data set for updating the posterior of the surrogate model. This process is iterated until the maximum loss lock signal to interference ratio is reached in the test or the condition that the signal to interference ratio cannot continue to increase is found in the preset search space. This method can effectively utilize a small amount of test evaluation to obtain key information, avoid blind test, and significantly reduce the number of tests.

[0107] In one embodiment, the specific implementation process of step S103 includes:

[0108] The multiple interference signal parameter sets in the current observation data set are input into the surrogate model at the same time, and the posterior characteristic parameters of the Gaussian process for the signal to interference ratio threshold prediction values of each interference signal parameter set are determined.

[0109] In the electromagnetic sensitivity test under multi-parameter electromagnetic interference in the present embodiment, the search space of electromagnetic interference parameters is often high-dimensional. On the one hand, the continuous high-dimensional search space leads to a longer operation time of algorithm search, and on the other hand, it is difficult to converge within a small number of test times. Therefore, it is necessary to set a reasonable search grid to discretize the parameter space, so as to avoid the curse of dimensionality caused by continuous parameter search. In addition, by setting discrete rather than continuous test parameters, the search space of electromagnetic interference potential parameters can also be artificially set by expert experience.

[0110] In order to improve the test efficiency, a multi-batch parallel test method is introduced in the learning electromagnetic sensitivity test design based on Bayesian optimization, that is, the batch number of each test can be adjusted according to the test grid. In addition, Bayesian optimization follows a repeatable path, which means that it will generate the same results at the same starting point.

[0111] The following two actual examples are used to verify the feasibility of the unmanned aerial vehicle data link complex electromagnetic interference sensitive parameter extraction method provided in the present embodiment.

[0112] The test examples under two typical interferences are: one is microwave pulse single-source interference with discontinuous waveform, and the other is double-source interference with continuous waveform, including continuous wave and wideband white Gaussian noise (White Gaussian Noise. WGN). The test is configured according to Figure 2 The working signal center frequency is f0MHz and the working power is set to -96.7dBm.

[0113] The electromagnetic sensitivity test design based on Bayesian optimization is carried out based on the Edbo package in Python3.7. In the test example, the application effects of PI, EI and UCB three classical acquisition functions are compared, and the electromagnetic interference parameter combination that is most susceptible to the data link is analyzed.

[0114] Example 1: Microwave pulse single-source interference with discontinuous waveform

[0115] Microwave pulse has been proved in research to have potential impact on the avionics system of unmanned aerial vehicle. Based on the unmanned aerial vehicle data link complex electromagnetic interference sensitive parameter extraction method, the interference effect of high repetition frequency microwave pulse on data link is studied. The parameters of microwave pulse include pulse repetition frequency (Pulse Repetition Frequency, PRF), duty cycle (Duty Cycle, DC) and carrier frequency. In the present embodiment, square wave modulation continuous wave is selected as the modulation mode of microwave pulse, which has multiple harmonics and is easy to realize high repetition frequency and duty cycle. In the test, the search range of duty cycle is set to T DC∈[20%, 30%...80%], the search range of pulse repetition frequency is f PRF ∈[1MHz,2MHz…10MHz]. Based on the performance of the data link during electromagnetic susceptibility testing under continuous wave interference, the search range of the carrier frequency fc is set to [f0-5MHz,f0-4MHz,…,f0+5MHz]. This embodiment aims to obtain the maximum signal-to-interference ratio threshold when the data link loses lock with the fewest number of tests possible, thereby finding a microwave pulse interference signal parameter combination that makes the data link most sensitive.

[0116] The experiment is conducted within the set parameter search range. The signal-to-interference ratio convergence curve obtained by the method provided in this embodiment is as follows: Figure 4 As shown. The initial test is randomly generated, and 10 groups of interference signal parameter groups are tested in each batch. The Bayesian algorithm is used to search for the global maximum value of the signal-to-interference ratio threshold. It can be concluded from the test results that under the guidance of the learning test method, the maximum value of the data link unlock signal-to-interference ratio obtained in each batch of tests gradually increases. It can be seen that the learning test method can effectively guide the search for the maximum value of the objective function. In the case of high-repetition-rate microwave pulse interference in the data link in this embodiment, the convergence speed of the adaptive test is relatively fast. In particular, when EI is used as the acquisition function, the received signal-to-interference ratio reaches the maximum value after the third batch of tests, and the received signal-to-interference ratio of PI and UCB reaches the maximum value after the fourth batch of tests. The EI acquisition function has the highest search efficiency in this case. Therefore, this embodiment preferably uses the EI function as the acquisition function to select the next test point.

[0117] Depend on Figure 4 As can be seen, if testers lack experience with the microwave pulse parameters that are most susceptible to data link interference, the learning-based electromagnetic sensitivity test method only requires 30 sets of tests to obtain the maximum signal-to-interference ratio threshold that causes data link lock to fail. Traditional electromagnetic sensitivity testing requires 7 × 10 × 11 = 770 sets of tests to reliably obtain the maximum signal-to-interference ratio threshold for lock failure. Compared to traditional sensitivity testing methods, the learning-based electromagnetic sensitivity test is approximately 26 times faster.

[0118] In the 30 sets of electromagnetic sensitivity tests conducted, the five maximum signal-to-interference ratio thresholds and their corresponding interference signal parameters were compared, as shown in Table 1. Within the set repetition frequency and duty cycle search range, when the repetition frequency of the microwave pulse signal is f PRF =1MHz, duty cycle T DC =80%, carrier frequency f c When the data link is close to the center frequency f of the data link working signal, the data link is more susceptible to interference. c=f0+1MHz, the signal-to-interference ratio threshold when the data link loses lock is the largest, which means that the data link is most sensitive to the interference signal parameter combination of the microwave pulse at this time.

[0119] Table 1

[0120] Carrier frequency (MHz) Duty cycle Repetition frequency (MHz) Signal to noise ratio [cd] f0+1 80％ 1 -13.6 f0-1 80％ 1 -14.1 f0 80％ 1 -15.6 <![CDATA[f0+2]]> 80％ 1 -19.3 [ f0-2 ] 80％ 1 -20

[0121] According to the microwave pulse parameter combination that makes the data link most susceptible in Table 1, the signal source output is set to f c =f0+1MHz, T DC =80%, f RPF =1MHz, power is -8dBm. The interference signal spectrum measured by the spectrum analyzer is as follows Figure 5 As shown in the figure, the shaded area is the passband of the IF filter of the RF front end of the data link receiver. It can be seen that the microwave pulse signal has multiple frequencies. There are eight frequency components in the receiver's frequency band, which may interfere with the data link.

[0122] ①RF front end

[0123] The signal received by the antenna includes the data link working signal and the electromagnetic interference signal in the space. After filtering, mixing and automatic gain control (AGC), the intermediate frequency signal yIF with stable power is output. Since the passband range of the intermediate frequency filter is [f0-5MHz, f0+5MHz], when the frequency of the interference signal falls within the passband range of the filter, the intermediate frequency output signal yIF IF May contain some in-band interference.

[0124] In order to analyze the effect of high repetition rate microwave pulse interference on the data link, the microwave pulse parameter combination that makes the data link most sensitive in Table 1 is used to interfere with the data link. Under normal working conditions, the intermediate frequency signal y IF The spectrum is an envelope with a bandwidth of 10 MHz. In order to observe the impact of the interference signal on the intermediate frequency signal, the single carrier power of the data link transmission is set to be consistent with the in-band power of the modulation signal, and the intermediate frequency signal waveform without interference is obtained as follows Figure 6 As shown in (a) of the figure. Due to the frequency conversion effect of the mixer, the carrier signal frequency is shifted to an intermediate frequency centered at 140MHz. In addition, under the action of the multi-stage amplifier and AGC, the in-band noise and the working signal carrier are amplified together, where the power of the working signal is 6.6dBm. When the data link is interfered with and loses lock, f RPF =1MHz, T DC =80%, f c=f0+1MHz microwave pulse interference when the data link loses lock, the intermediate frequency signal is as follows Figure 6 As shown in (b) of the figure, it can be seen that multiple frequency components generated by the interfering signal can pass through the receiver's RF front-end. Since the AGC amplifier circuit cannot distinguish between the working signal and the interfering signal, it can only adjust the gain based on the total strength of the input signal. As a result, the harmonic component at a frequency of 140 MHz has a significant power suppression effect on the working signal, reducing the signal power at the working signal frequency by 7.8 dB.

[0125] ②Data terminal analysis

[0126] Intermediate frequency signal y IF After sampling by the Analog-to-Digital Converter (ADC), the continuous analog signal is converted into a baseband discrete digital signal. Figure 7 As shown in Figure 1, the baseband signal is despread at position A and then passes through the demodulator and Viterbi decoder. Frame synchronization is performed at position B. Whether frame synchronization can be achieved determines whether the data link is unlocked.

[0127] Before the received signal is correctly despread, the local pseudo code generator Figure 7 At point B in the spread spectrum, a local pseudo code is generated that is perfectly aligned with the received pseudo code phase. This local code changes with the received pseudo code. This process is called spread spectrum pseudo code synchronization. Pseudo code synchronization involves synchronization acquisition and synchronization tracking. Acquisition is the first critical and time-consuming step in the receiver's digital processing. If the phase search fails, it means that the useful signal has not been captured, and subsequent demodulation and decoding of the signal are meaningless.

[0128] In pseudo-code synchronization, the time required to enter the locked state from the initial state or unlocked state when the acquisition loop receives an external signal is called the acquisition time. Calculated based on the code length, information rate, and accumulation time, the data link acquisition time is 320ms. The high-repetition-rate microwave pulse width studied in the examples is much shorter than the loop's acquisition time. Therefore, once interference causes the data link to lose lock, the loop cannot complete the acquisition process again. Furthermore, at a given pulse repetition frequency, the higher the duty cycle, the greater the power of the interfering signal entering the acquisition loop within the same timeframe. Therefore, data links are more sensitive to high-duty-cycle pulsed interference signals.

[0129] In the parameter search range of high repetition frequency microwave pulse interference, by analyzing the electromagnetic interference parameters screened by the learning type electromagnetic sensitivity test provided in the embodiment, it can be concluded that: (1) the pulse repetition frequency and the carrier frequency will have a joint effect on the data link, which is specifically manifested in that multiple harmonics generated by the repetition frequency and the carrier frequency modulation fall within the sensitive frequency band of the data link; (2) due to the unconditional amplitude stabilization effect of the AGC amplifier on the signal, the gain of the useful signal is compressed; (3) the high repetition frequency signal with a pulse width less than the capture time makes it difficult for the capture loop to recover from the out-of-lock state, and therefore, compared with the low duty cycle pulse modulation signal, the high duty cycle will cause more serious blocking effect in the capture loop.

[0130] Example Two: Dual-source electromagnetic interference test

[0131] The dual-source interference is set as a continuous wave and a wideband Gaussian white noise signal, both of which are commonly used suppression type interference signals in electromagnetic sensitivity tests. The continuous wave signal is the most widely used interference type in the military standard electromagnetic sensitivity test, and the frequency is the only parameter affecting the sensitivity threshold. The wideband Gaussian white noise signal is widely used in anti-UAV equipment and has two variable parameters of center frequency and bandwidth, which is referred to as wideband white noise hereinafter. The superposition of the continuous wave and the wideband white noise is a typical case of complex electromagnetic environment, which can interfere with the sensitive frequency of the data link and suppress the key frequency band at the same time.

[0132] The test is set according to Figure 3 The two interference signals are combined through a combiner and then input to the radio frequency front end to interfere with the data link. The method for extracting the complex electromagnetic interference sensitivity parameters of the UAV data link provided in the embodiment is used to search the combination of multiple electromagnetic interference source signal parameters. The goal of the sensitivity test is to make the signal-to-interference ratio of each interference signal when the data link is out of lock as large as possible, so as to search for the electromagnetic interference parameter combination that makes the data link most sensitive when the two interference sources exist at the same time.

[0133] Specifically, the dual-source test process is as follows:

[0134] According to the sensitive parameter range of the data link to the continuous wave interference, the search range of the continuous wave interference frequency f1 and the wideband white noise interference frequency f2 is set as [f0-5MHz, f0+5MHz]. The bandwidth f Bw of the wideband white noise signal is set as [1MHz, 10MHz] with a step interval of 1MHz. When the continuous wave and the wideband white noise act on the data link respectively, the signal-to-interference ratios of the antenna received signals when out of lock are represented as R1' and R2' respectively; when the two interference sources act on the data link at the same time, the signal-to-interference ratios of the antenna received signals when out of lock are represented as R1 and R2.

[0135] In the embodiment, the search target function of the unmanned aerial vehicle data link complex electromagnetic interference sensitive parameter extraction method is set as the sum R1+R2 of the continuous wave and Gaussian white noise signal-to-interference ratios when the data link is out of lock. Although the sum of the signal-to-interference ratios has no physical meaning, when the value reaches the highest, it can represent the case of the best interference effect under the joint action of the two interference sources.

[0136] In the case of a working signal strength of -96.7 dBm, the electromagnetic interference test of the unmanned aerial vehicle data link by the continuous wave and wideband white noise is carried out, and the specific test process is as shown in Figure 8 The unmanned aerial vehicle data link complex electromagnetic interference sensitive parameter extraction method based on Bayesian optimization is embedded between the single-source interference test and the double-source interference test, and plays a role in guiding the selection of the interference source parameters in the test.

[0137] Under the condition of the randomly initialized interference signal parameter group of the test, 10 groups of double-source interference signal parameter combinations are selected for each batch for the test. The Bayesian optimization algorithm is used to search the test results of the initialized double-source interference signal parameter combinations, and after training the Gaussian process proxy model, new test points are selected in the preset search space by optimizing the acquisition function until the maximum sum of the out-of-lock signal-to-interference ratios is reached. The convergence curves obtained by using different acquisition functions for the electromagnetic sensitivity test are as shown in Figure 9 It can be concluded that the sum of the out-of-lock signal-to-interference ratios reaches the maximum value in the 4th test when the acquisition function is EI and PI, and the convergence speed is the slowest when the acquisition function is UCB, which reaches the maximum value only after the 7th iteration test. By comparing the signal-to-interference ratio convergence curves under the three acquisition functions, it can be concluded that the convergence of the test is the best when the acquisition function is EI. Without experience reference, the most sensitive electromagnetic interference parameter combination can be obtained after the 40th test by using the method provided in the embodiment, while the sensitive parameter combination obtained by using the traditional test method requires 10x11x11=1210 tests.

[0138] The electromagnetic sensitivity thresholds of the unmanned aerial vehicle data link under the PI, EI and UCB acquisition functions are obtained according to the method of the embodiment, and five groups of interference source parameters that make the data link most susceptible to influence are extracted, as shown in Table 2. In the case of simultaneous interference of the continuous wave and the wideband white noise, the sensitive frequencies of the continuous wave and the wideband noise are both near the working center frequency f0, wherein, when the continuous wave frequency is f 0+1 MHz, the wideband noise center frequency is f0MHz, and the bandwidth is 2MHz, the data link is most susceptible to interference.

[0139] Table 2

[0140]

[0141] In order to analyze the relationship between the combination of double-source electromagnetic interference sensitive parameters and single-source interference sensitive parameters leading to the data link losing lock, the single-source interference test is performed on the data link by using continuous wave and wideband white noise respectively. As shown in Figure 10 , the lock loss signal-to-interference ratio threshold of the data link under continuous wave and wideband white noise interference is basically symmetrical with the working center frequency of the data link. When the continuous wave interference frequency is f0+1MHz, the maximum lock loss signal-to-interference ratio of the data link is-12.2dB. For different bandwidths of the wideband white noise interference, the electromagnetic sensitivity threshold of the data link is tested within the center frequency range [f0-5MHz, f0MHz], and the result is shown in Figure 11 . When the center frequency of the wideband white noise interference is f0MHz and the bandwidth is 2MHz, the maximum signal-to-interference ratio of the data link is-13.6dB.

[0142] By using continuous wave and wideband white noise to interfere with the data link, the feasibility of the learning electromagnetic sensitivity test method in double-source electromagnetic interference test is verified. When two interference sources interfere with the data link at the same time, the sensitive frequencies of the continuous wave and wideband white noise interference are f0+1MHz and f0MHz respectively, and the sensitive bandwidth of the wideband white noise interference is 2MHz, and these interference source parameters are also the single-source interference parameters that are easy to make the data link electromagnetic sensitive.

[0143] From the above embodiment, it can be seen that the maximum signal-to-interference ratio threshold of the data link when losing lock is taken as the search target of the sensitivity, and the unmanned aerial vehicle data link complex electromagnetic interference sensitive parameter extraction method based on Bayesian optimization is established. The application in the cases of high repetition frequency microwave pulse interference and double-source interference shows that the convergence effect of the electromagnetic sensitivity test is better when the EI acquisition function is applied, and the learning electromagnetic sensitivity test method can accelerate about 26 times and 30 times respectively compared with the traditional uniform test method.

[0144] Secondly, the sensitive parameter distribution rule of the high repetition frequency microwave pulse interference is determined. When the microwave pulse interference signal has a high duty cycle, the carrier frequency is close to the center frequency f0 of the data link working signal, and the repetition frequency makes the interference harmonics more fall within the data link passband, the data link is more sensitive to the microwave pulse interference. The reason is that when the harmonics generated by the pulse repetition frequency and the carrier frequency modulation fall within the sensitive frequency band of the data link, the gain compression of the working signal is caused, and the blocking effect in the data link capture loop caused by the high duty cycle interference signal is difficult to recover.

[0145] In addition, the method provided by the embodiment provides a dual-source electromagnetic interference sensitivity test framework, takes the sum of the signal-to-interference ratios of the two interference sources in a lock loss state as a search target of an algorithm, and obtains a parameter combination of the dual-source electromagnetic interference most sensitive to the data link through the test. When the continuous wave is a near-frequency interference, the center frequency of the broadband white noise interference is the same as the center frequency f0 of the data link working signal, and the bandwidth is near 2MHz, the data link is more susceptible to interference, and the above sensitive parameters are still valid when the continuous wave and the broadband white noise are used as single-source interference and act on the data link.

[0146] In summary, the unmanned aerial vehicle data link complex electromagnetic interference sensitive parameter extraction method provided by the embodiment is an efficient data-driven method, which can more quickly obtain an electromagnetic sensitive parameter combination, reduce the manpower, material resources and time cost of electromagnetic sensitivity test, and lay a foundation for the effect research of the data link under complex electromagnetic interference signals. Moreover, the method provided by the embodiment can be widely applied to electromagnetic sensitivity test of any tested device, and can be extended to multiple interference sources from two interference sources.

[0147] It should be understood that the size of the serial number of each step in the above embodiment does not mean the order of execution, and the execution order of each process should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiment of the application.

[0148] The following is a device embodiment of the application, and for details not described in detail, reference can be made to the corresponding method embodiments described above.

[0149] The unmanned aerial vehicle data link complex electromagnetic interference sensitive parameter extraction device provided by the embodiment of the application comprises a structure, which comprises:

[0150] An observation data set initialization module is configured to initialize an observation data set, wherein the observation data set comprises at least one observation data, the observation data comprises a group of interference signal parameters and a signal-to-interference ratio threshold value of the data link when the unmanned aerial vehicle loses lock under electromagnetic interference of the group of interference signal parameters, and the group of interference signal parameters comprises a plurality of electromagnetic interference signal parameters.

[0151] A proxy model construction module is configured to construct a target function by taking the group of interference signal parameters as independent variables and maximizing the signal-to-interference ratio threshold value when the data link loses lock as an optimization target, and construct a proxy model of the target function by using a Gaussian process.

[0152] A posterior characteristic parameter acquisition module is configured to input the group of interference signal parameters in the current observation data set into the proxy model, and determine a posterior characteristic parameter of an observed value of the signal-to-interference ratio threshold value of the group of interference signal parameters by the Gaussian process; the posterior characteristic parameter comprises a mean value and / or a variance.

[0153] The observation set updating module is configured to construct a collection function based on the posterior characteristic parameter, select a group of interference signal parameters as a next test point in the search space by maximizing the collection function to perform the electromagnetic interference experiment, determine a signal-to-interference ratio threshold corresponding to the next test point, and store the next test point and the corresponding signal-to-interference ratio threshold as a group of observation data in a current observation data set.

[0154] The interference signal parameter group output module is configured to repeatedly perform the above steps until a preset convergence condition is met, and output a group of interference signal parameters corresponding to a maximum signal-to-interference ratio threshold at which the data link is lost.

[0155] In one possible implementation, the observation data set initialization module includes:

[0156] a plurality of groups of interference signal parameters are randomly generated in the parameter search space, and an electromagnetic interference experiment on the unmanned aerial vehicle is performed on each group of interference signal parameters by using an electromagnetic interference injection test method to determine a signal-to-interference ratio threshold at which the data link of the unmanned aerial vehicle is lost under electromagnetic interference of each group of interference signal parameters;

[0157] Based on the formula y i =R(x i )+ε i , i=1, …, N, an observation value corresponding to each group of interference signal parameters is determined, a group of observation data composed of the interference signal parameters and the corresponding observation value is generated, and an observation data set is generated.

[0158] wherein R(x i ) represents the signal-to-interference ratio threshold, ε i represents additive noise with a distribution of , x i represents the i th group of interference signal parameters in the observation data set, and y i represents the observation value of the i th group of interference signal parameters.

[0159] In one possible implementation, the proxy model construction module includes:

[0160] The mean function and the kernel function of the Gaussian process are set as the prior distribution, wherein the Gaussian process is defined as:

[0161] R(x) ~ GP(m(x), k θ (x i , x j ));

[0162] wherein R(x) represents a regression function, m(x) represents a mean function and is set to zero; k θ (x i , x j ) represents a kernel function, and a denotes an output scaling factor, x i and x j denote the i-th and j-th interference signal parameter set in the observation data set, respectively; r = ||x i -x j ||, l denotes a characteristic length scale;

[0163] For an input value x of the Gaussian process, the regression function and the observation value follow a joint Gaussian prior distribution:

[0164]

[0165] where k θ (x, X) = [k θ (x, x1), k θ (x, x2), …, k θ (x, x N )].

[0166] The conditional posterior distribution of R(x) is: where μ(x) denotes the mean of the Gaussian process, σ 2 (x) denotes the variance of the Gaussian process; and

[0167]

[0168] where X denotes the set of interference signal parameter sets; K denotes the N x N covariance matrix between input values, D denotes the observation data set, y denotes the observation value, and I denotes the identity matrix.

[0169] In one possible implementation, the acquisition function comprises a modified probability acquisition function:

[0170] The expression of the modified probability acquisition function is:

[0171] where f(x) denotes the mapping relationship between the interference signal parameter set and the signal-to-interference ratio threshold, x + denotes the current optimal value of the objective function, μ(x) denotes the mean of the Gaussian process, σ 2 (x) denotes the variance of the Gaussian process; Φ(·) denotes the cumulative distribution function, and D denotes the observation data set.

[0172] In one possible implementation, the acquisition function comprises an expected improvement acquisition function;

[0173] The expression of the expected improvement acquisition function is: where φ(z) denotes the probability density function; f(x) denotes the mapping relationship between the interference signal parameter set and the signal-to-interference ratio threshold, x +wherein x represents the current optimal solution of the objective function, μ(x) represents the mean of the Gaussian process, and σ(x) represents the standard deviation of the Gaussian process.

[0174] In one possible implementation, the apparatus provided by the embodiment further includes a model training module configured to:

[0175] estimate the hyperparameters of the proxy model by maximizing a log marginal likelihood function;

[0176] optimize the hyperparameters of the proxy model by iteratively optimizing the log marginal likelihood function using a gradient descent method.

[0177] In one possible implementation, the log marginal likelihood function is expressed as:

[0178]

[0179] wherein θ represents a set of hyperparameters, and X represents a set of interference signal parameter groups, represents a noise variance, l represents a characteristic length scale, a represents an output scale factor, and N represents the number of observation data in the observation data set; K represents an N x N covariance matrix between input values, and I represents an identity matrix.

[0180] The embodiment of the present application further provides an electronic device including a memory and a processor, the memory stores a computer program, and the processor implements the method in the above method embodiment when executing the computer program. Exemplarily, the electronic device can be a desktop computer, a notebook computer, a palm computer, a cloud server, and the like, without any limitation.

[0181] In the above embodiments, the description of each embodiment has its own focus, and the parts not described or recorded in a certain embodiment can be referred to the relevant description of other embodiments. If there is no special description and no logical conflict, the terms and / or descriptions of different embodiments are consistent and can be mutually referenced. The technical features in different embodiments can be combined to form a new embodiment according to their inherent logical relationship.

[0182] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than limit them; even though the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A method for extracting complex electromagnetic interference sensitive parameters of UAV data link, characterized by: include: Step 1: Initialize an observation data set, wherein the observation data set includes at least one observation data, the observation data including an interference signal parameter group and a signal-to-interference ratio threshold value when the UAV loses data link lock under electromagnetic interference of the interference signal parameter group; the interference signal parameter group includes multiple electromagnetic interference signal parameters; Step 2: Using the interference signal parameter group as the independent variable and maximizing the signal-to-interference ratio threshold when the data link is locked as the optimization goal, an objective function is constructed, and a surrogate model of the objective function is constructed using a Gaussian process. Step 3: Input the interference signal parameter group in the current observation data set into the proxy model to determine the posterior characteristic parameters of the observation value of the signal-to-interference ratio threshold of the Gaussian process for the interference signal parameter group; the posterior characteristic parameters include mean and / or variance; Step 4: Construct an acquisition function based on the a posteriori characteristic parameters, and select an interference signal parameter group as the next test point in the search space by maximizing the acquisition function to perform an electromagnetic interference experiment, determine the signal-to-interference ratio threshold corresponding to the next test point, and store the next test point and the corresponding signal-to-interference ratio threshold as a set of observation data in the current observation data set; Step 5: Return to step 3 and repeat until the preset convergence condition is met, and output the interference signal parameter group corresponding to the maximum signal-to-interference ratio threshold when the data link is lost.

2. The method for extracting complex electromagnetic interference sensitive parameters of UAV data link according to claim 1 is characterized in that: The initialization of the observation data set includes: Multiple interference signal parameter groups are randomly generated in the parameter search space, and electromagnetic interference tests are conducted on each interference signal parameter group using the electromagnetic interference injection test method to determine the signal-to-interference ratio threshold at which the drone loses data link lock under electromagnetic interference of each interference signal parameter group. Based on the formula y i =R(x i )+ε i , i=1,…,N determine the observation value corresponding to each interference signal parameter group, combine the interference signal parameter group and the corresponding observation value into a set of observation data, and generate an observation data set; Among them, R(x i ) represents the signal-to-interference ratio threshold, ε i It means that the distribution is The additive noise, x i represents the i-th interference signal parameter group in the observation data set, y i represents the observed value of the i-th interference signal parameter group.

3. The method for extracting complex electromagnetic interference sensitive parameters of UAV data link according to claim 1 is characterized in that: The specific process of constructing the proxy model in step 2 includes: Set the mean function and kernel function of the Gaussian process as the prior distribution; the Gaussian process is defined as: Among them, R(x) represents the regression function, m(x) represents the mean function, which is set to zero; k θ (x i ,x j ) represents the kernel function, and a represents the output scale factor, x i and x j Respectively represent the i-th interference signal parameter group and the j-th interference signal parameter group in the observation data set; r=||x i -x j ||, l represents the characteristic length scale; For the input value x of the Gaussian process, the regression function and the observation value follow a joint Gaussian prior distribution: where k θ (x,X) = [k θ (x,x1), k θ (x,x2), …, k θ (x,x N )]; The conditional posterior distribution of R(x) is: Among them, μ(x) represents the mean of the Gaussian process, σ 2 (x) represents the variance of the Gaussian process; and Where X represents the set of interference signal parameters; K represents the N×N covariance matrix between input values, D represents the observation data set, y represents the observation value, and I represents the identity matrix; k θ (x, X) represents the covariance vector between the newly input interference signal parameter group and the interference signal parameter group set; k θ (x,x N ) represents the covariance between the newly input interference signal parameter group and the Nth interference signal parameter group in the interference signal parameter group set.

4. The method for extracting complex electromagnetic interference sensitive parameters of UAV data link according to claim 1 is characterized in that: The acquisition function includes a desired improved acquisition function; The expression of the expected improved acquisition function is: in, φ(z) represents the probability density function; f(x) represents the mapping relationship between the interference signal parameter group and the signal-to-interference ratio threshold, x + represents the current optimal solution of the objective function, μ(x) represents the mean of the Gaussian process, and σ(x) represents the standard deviation of the Gaussian process.

5. The method for extracting complex electromagnetic interference sensitive parameters of UAV data link according to claim 1 is characterized in that: The acquisition function includes an improved probability acquisition function: The expression of the improved probability acquisition function is: Among them, f(x) represents the mapping relationship between the interference signal parameter group and the signal-to-interference ratio threshold, x + represents the current optimal value of the objective function, μ(x) represents the mean of the Gaussian process, σ 2 (x) represents the variance of the Gaussian process; Φ(·) represents the cumulative distribution function, and D represents the observed data set.

6. The method for extracting complex electromagnetic interference sensitive parameters of UAV data link according to claim 1 is characterized in that: The method further comprises: estimating the hyperparameters of the surrogate model by maximizing the log-marginal likelihood function; The log-marginal likelihood function is iteratively optimized by a gradient descent method to optimize the hyperparameters of the surrogate model.

7. The method for extracting complex electromagnetic interference sensitive parameters of UAV data link according to claim 6 is characterized in that: The expression of the log-marginal likelihood function is: where θ represents the set of hyperparameters, and X represents the interference signal parameter set, represents the noise variance, l represents the characteristic length scale, a represents the output scale factor, N represents the number of observations in the observation dataset; K represents the N×N covariance matrix between the input values, and I represents the identity matrix.

8. A device for extracting complex electromagnetic interference sensitive parameters of UAV data link, characterized in that: include: an observation data set initialization module, configured to initialize an observation data set, the observation data set including at least one observation data, the observation data including an interference signal parameter group and a signal-to-interference ratio threshold value that causes the UAV to lose data link lock under electromagnetic interference of the interference signal parameter group; the interference signal parameter group including multiple electromagnetic interference signal parameters; a proxy model construction module, configured to construct an objective function using an interference signal parameter group as an independent variable and maximizing a signal-to-interference ratio threshold when the data link is locked as an optimization goal, and to construct a proxy model of the objective function using a Gaussian process; a posterior characteristic parameter acquisition module, configured to input the interference signal parameter group in the current observation data set into the proxy model, and determine the posterior characteristic parameters of the observation value of the signal-to-interference ratio threshold of the Gaussian process for the interference signal parameter group; the posterior characteristic parameters include mean and / or variance; an observation set updating module, configured to construct an acquisition function based on the posterior characteristic parameters, select an interference signal parameter group as the next test point in the search space by maximizing the acquisition function, determine a signal-to-interference ratio threshold corresponding to the next test point, and store the next test point and the corresponding signal-to-interference ratio threshold as a set of observation data in the current observation data set; The interference signal parameter group output module is used to repeatedly execute the above steps until the preset convergence condition is met, and output the interference signal parameter group corresponding to the maximum signal-to-interference ratio threshold when the data link is lost.

9. An electronic device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images