Cognitive radar identification method and system based on K-S test and momentum gradient descent

Abstract

The embodiment of the invention provides a radar identification method and system based on K-S test and momentum gradient descent, and relates to the technical field of electronic warfare signal processing, and the method comprises the steps: estimating the related parameters of a radar signal according to a filtered available echo signal, deducing the distribution of a signal to interference plus noise ratio (SINR), and carrying out the estimation of the related parameters of the radar signal; obtaining radar SINR sample distribution and pre-deduced theoretical signal and interference plus noise ratio SINR distribution; according to radar SINR sample distribution and pre-deduced theoretical signal and interference plus noise ratio SINR distribution, K-S test is used to sense the adaptive power distribution capability of the radar, and a momentum gradient descent method is used to optimize interference signals. The result shows that the K-S detector has better performance on recognition of the cognitive radar and the non-cognitive radar, and when the probability of the second-class error is reduced to 1%, the performance of the K-S detector is improved by about 10 db compared with that of an Afriat detector.

Description

Technical Field

[0001] The present invention relates to the technical field of electronic warfare signal processing, and in particular to a cognitive radar recognition method and system based on KS test and momentum gradient descent. Background Art

[0002] As modern warfare enters the information age, the importance of information on the battlefield has increased unprecedentedly, and mastering the information initiative has become the key to victory. As an important part of information warfare, radar electronic countermeasures are becoming increasingly prominent in information warfare. In the field of radar electronic countermeasures (ECM), drones need to identify the working status of target radars and select interference measures accordingly. For ordinary radars, current research focuses on using machine learning methods to identify unknown radar states. Supervised learning methods include artificial neural networks, decision trees, K nearest neighbor algorithms, and support vector machines (SVMs).

[0003] However, with the emergence of cognitive radar, the characteristic of realizing optimal autonomous decision-making and adjustment of waveform parameters through perception and cognition of the surrounding environment or targets makes it difficult to identify the state of traditional radar. At present, there are few studies on cognitive radar recognition. Some scholars have proposed a cognitive electronic countermeasure (CECM) architecture, integrating cognitive architecture into radar electronic countermeasure system, and providing a macro framework for cognitive radar countermeasure system. Some researchers have also used convolutional neural network (CNN) to classify the parameters of cognitive radar descriptors to identify whether the radar has cognitive characteristics, and some scholars have used Afriat's theorem in microeconomics to identify whether the radar has cognitive behavior. However, the current methods lack the recognition of certain specific functions of cognitive radar, such as optimizing power resource allocation, dynamic spectrum management or space resource management. And as an emerging field, cognitive radar recognition has great research potential. Summary of the invention

[0004] In view of the problem that there are few cognitive radars and cognitive radar specific function identification methods at present, the present invention proposes a technical solution for cognitive radar identification based on KS test to identify cognitive radars with adaptive power allocation function. The technical solution mainly includes that the UAV changes the emitted interference signal and collects the corresponding radar signal, then estimates the power of the radar signal and derives the distribution of the signal to interference plus noise ratio (SINR), obtains the radar SINR sample distribution and the pre-derived theoretical SINR distribution, uses the KS test to perceive the radar's adaptive power allocation capability, and uses the momentum gradient descent method to optimize the interference signal to improve the performance of the KS (Kolmogorov-Smirnov) detector. Finally, the performance simulation of the KS detector and the comparative test of the KS detector and the Afriat theorem detector are carried out.

[0005] To achieve the above object, in a first aspect, the present invention proposes a radar identification method based on KS test and momentum gradient descent, comprising:

[0006] Perform interference filtering on the original echo signal collected by the radar of the UAV to obtain a usable echo signal after filtering; the original echo signal includes: an interference signal spectrum; the usable echo signal includes: a radar signal spectrum;

[0007] According to the available echo signal after filtering, the relevant parameters of the radar signal are estimated and the distribution of the signal to interference plus noise ratio SINR is derived, and the radar SINR sample distribution and the pre-derived theoretical signal to interference plus noise ratio SINR distribution are obtained;

[0008] According to the radar SINR sample distribution and the pre-derived theoretical signal to interference plus noise ratio SINR distribution, the KS test is used to perceive the radar's adaptive power allocation capability, and the momentum gradient descent method is used to optimize the interference signal.

[0009] The original echo signal collected by the radar of the UAV is subjected to interference filtering, and the filtered available echo signal includes:

[0010] Performing frequency domain analysis on the original echo signal collected by the radar of the UAV to obtain relevant parameters of the original echo signal; the relevant parameters of the original echo signal include: at least one of noise power density, interference signal spectral density, and target signal spectral density;

[0011] An adaptive power allocation algorithm is used to optimize the original echo signal according to the noise power density, interference signal spectral density and target signal spectral density.

[0012] The relevant parameters of the original echo signal also include: noise power density, interference signal spectral density, target signal spectral density;

[0013] The distribution of signal to interference plus noise ratio SINR is derived based on the noise power density, interference signal spectral density and target signal spectral density.

[0014] According to the available echo signal after filtering, the relevant parameters of the radar signal are estimated and the distribution of the signal to interference plus noise ratio SINR is derived, and the radar SINR sample distribution and the pre-derived theoretical signal to interference plus noise ratio SINR distribution are obtained, including: using a KS detector, calculating the test statistic and the threshold under a given significance level τ.

[0015] According to the radar SINR sample distribution and the pre-derived theoretical signal to interference plus noise ratio SINR distribution, the KS test is used to perceive the radar's adaptive power allocation capability, and the momentum gradient descent method is used to optimize the interference signal, including: using the momentum gradient descent algorithm to optimize the jammer's signal to reduce the probability of Type II error in radar identification, and comparing it with a cognitive radar detector based on the economic Afriat theorem.

[0016] In a second aspect, the present invention proposes a radar identification device based on KS test and momentum gradient descent, comprising:

[0017] The interference filtering module is used to perform interference filtering on the original echo signal collected by the radar of the UAV to obtain a usable echo signal after filtering; the original echo signal includes: an interference signal spectrum; the usable echo signal includes: a radar signal spectrum;

[0018] An estimation module is used to estimate the relevant parameters of the radar signal and derive the distribution of the signal to interference plus noise ratio SINR according to the available echo signal after filtering, and obtain the radar SINR sample distribution and the pre-derived theoretical signal to interference plus noise ratio SINR distribution;

[0019] The test module is used to sense the adaptive power allocation capability of the radar using the KS test according to the radar SINR sample distribution and the pre-derived theoretical signal to interference plus noise ratio SINR distribution, and optimize the interference signal using the momentum gradient descent method.

[0020] In a third aspect, the present invention proposes an electronic device, comprising: a processor and a memory, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor executes the machine-readable instructions to execute the steps of the radar identification method based on KS test and momentum gradient descent as described above.

[0021] In a fourth aspect, the present invention proposes a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the radar identification method based on KS test and momentum gradient descent as described above are executed.

[0022] Compared with the prior art, the beneficial technical effects of the present invention are as follows:

[0023] The embodiment of the present invention proposes a method for identifying a cognitive radar with a power allocation function. The signal to interference plus noise ratio (SINR) distribution of the cognitive radar is estimated by the characteristic function method, and a KS detector is designed to use hypothesis testing to identify whether the target radar has the cognitive function of self-use power allocation. Then, a momentum gradient descent algorithm is used to optimize the jammer's signal to reduce the probability of type II error in radar identification, and compared with a cognitive radar detector based on the economic Afriat theorem. The results show that the KS detector has better performance in identifying cognitive radars and non-cognitive radars. When the probability of type II error is reduced to 1%, the performance of the KS detector is improved by about 10db compared with the Afriat detector. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the embodiments of the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0025] Figure 1 It is a schematic diagram of cognitive radar recognition provided by an embodiment of the present invention.

[0026] Figure 2 It is a radar signal spectrum and an interference signal spectrum provided by an embodiment of the present invention.

[0027] Figure 3 is the radar detection probability provided by the embodiment of the present invention.

[0028] Figure 4 It is a comparison diagram of an optimized signal and a non-optimized signal provided by an embodiment of the present invention.

[0029] Figure 5 It is a comparison diagram between the KS detector and the Afriat theorem detector provided by an embodiment of the present invention.

[0030] Figure 6 It is a schematic diagram of radar recognition based on KS test and momentum gradient descent provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0031] The exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present invention and to enable the scope of the present invention to be fully communicated to those skilled in the art.

[0032] like Figures 1 to 5As shown, an embodiment of the present invention provides a radar identification method based on KS test and momentum gradient descent, comprising:

[0033] Step 1: Perform frequency domain analysis on the radar signal, perform interference filtering on the original echo signal collected by the radar of the drone, and obtain a usable echo signal after filtering; the original echo signal includes: an interference signal spectrum; the usable echo signal includes: a radar signal spectrum.

[0034] Assuming that the target radar is a cognitive radar with optimized energy allocation function and uses matched filtering at the receiver end, the radar signal received by our jammer is:

[0035] x(t)=s(t)+n(t)

[0036] s(t) is the radar signal, n(t) is Gaussian noise, and n(t)~N(0,σ 2 ) Perform discrete Fourier transform DFT on the signal samples to obtain the frequency domain expression:

[0037] X(f)~N(S(f),σ 2 (f))

[0038] For random models of finite duration, the power spectral density is replaced by the energy spectral variance.

[0039] σ 2 (f) = E[|X(f)-S(f)| 2 ]=S xx (N(f))=σ 2

[0040] It is also necessary to estimate the radar signal power spectrum density S, which should satisfy the following expression under the total power constraint:

[0041]

[0042] P n (f i ) represents the noise power density, P d (f i ) is the spectral density of the interference signal sent by the jammer, P g (f i ) is the spectral density of the target signal detected by the radar. The total energy limit of the radar signal is E = ∫ Bw |S(f)| 2 Our jammer receives radar signals in noise and needs to estimate the total energy of the signal sent by the radar. Assuming E~N(μ,σ 2 ), and the moment estimation method is used to estimate:

[0043]

[0044] Generally speaking, the power spectrum of a random process is a definite function of the statistical average of samples. However, our method considers each sample realization in the random process without taking the statistical average of these samples. This is done to obtain the power spectrum of multiple samples for subsequent KS test. Therefore, the power spectrum of the sample obtained is a realization of the random process in the entire frequency band. The corresponding value of the radar signal power spectrum density at each frequency point is a random variable, which is the superposition of the true value of the radar signal power spectrum density and the frequency domain noise. We use the estimator Substitute the truth value |S(k)|.

[0045]

[0046] Step 2: Derive the distribution of signal SINR samples. According to the available echo signal after filtering, estimate the relevant parameters of the radar signal and derive the distribution of the signal to interference plus noise ratio SINR.

[0047] To facilitate subsequent derivation and representation, each frequency point index is replaced by i to obtain the SINR spectral density. The random variable corresponding to each frequency point i is:

[0048]

[0049] The random variable Y is a function of the random variable X. According to the definition of random variables, we can get:

[0050]

[0051] We can further simplify the formula:

[0052]

[0053] After we obtain the value Y of the SINR spectral density at each frequency point, the total SINR is obtained by discretizing the integral of the value at each frequency point on the frequency axis:

[0054]

[0055] Δf is the frequency resolution, that is, the smallest unit into which the frequency axis is divided. i (i=1,2,3...N) are independent of each other, which is equivalent to the result of superimposing independent Gaussian random noise on the true value of each frequency point of SINR spectrum density in the frequency domain. The characteristic function is used to solve the distribution of the sum of independent random variables, and then the characteristic function of random variable Yi is solved:

[0056]

[0057] Finally, the characteristic function of SINR at a certain moment when the radar sends the signal is obtained:

[0058]

[0059] The probability density function is obtained by inverting the characteristic function:

[0060]

[0061] Step 3: Construct the test statistic of the KS test to obtain the radar SINR sample distribution and the pre-derived theoretical signal to interference plus noise ratio SINR distribution.

[0062] We now propose the null hypothesis and alternative hypothesis as follows:

[0063]

[0064] Two error probabilities are given as follows:

[0065]

[0066] The L SINR sample matrices obtained are:

[0067] Z=[z 1 ,z 2 ,z 3 .....z L ] T

[0068] The obtained L sample observation distribution functions are as follows:

[0069]

[0070] According to the calculated theoretical SINR distribution of cognitive radar:

[0071]

[0072] Construct the test statistic:

[0073]

[0074] Arrange the collected radar SINR data from small to large to form an empirical distribution function. Then, according to the cognitive radar theoretical SINR distribution function, solve the difference between the SINR empirical distribution value of each sample and the corresponding theoretical distribution value. Find the value with the largest difference among the L sample points and compare it with the threshold γ. If it is greater than the threshold, the original hypothesis is rejected. If the difference is less than the threshold γ, we accept the original hypothesis.

[0075] Step 4: Given a significance level, perform a KS test, perceive the radar's adaptive power allocation capability, and use the momentum gradient descent method to optimize the jamming signal.

[0076] During the detection process, the spectrum of the interference signal sent is a random phase-modulated noise interference with different intensities applied to each sub-band:

[0077]

[0078] G sum (f) = ζ a G a (f)+ζ b G b (f)+ζ c G c (f)

[0079] ζ represents the weighting coefficient, G a (f) is the power spectrum density of noise phase modulation interference. The interference signal is sent M times, and the corresponding signal sent by the radar is:

[0080] α=[α 1 ,α 2 ,α 3 ,........α M ] T (i=1,2,3.....M)

[0081] β=[β 1 ,β 2 ,β 3 ,........β M ] T (i=1,2,3.....M)

[0082]

[0083] 1(.) Here is an indicator function, which means that it takes the value of 1 when (T≤γ) and 0 otherwise. is the detection probability. The jammer sends a jamming signal, and the signal sent for the Kth time is α i , remember we send α i The signal sent by the rear radar is β i The probability of correctly identifying cognitive radar and non-cognitive radar is increased by sending multiple jamming signals for probability accumulation.

[0084] The KS test gives a relevant analysis of the probability of correct detection and the decision threshold. According to the Kolgomolov criterion, for any λ>0, the following formula is obtained:

[0085]

[0086] When a significance level τ is given, it is also the upper bound of the first type of error, satisfying P{H1|H0}≤τ, and finding P{T>T τ}=τ’s critical value, which is the threshold γ we want to obtain.

[0087] Because the SINR samples in cognitive radar recognition are obtained under frequency domain noise σ, and the parameters of the theoretical distribution are estimated based on the data, when the noise is high, the threshold deviation is large. In order to obtain a more accurate threshold value, Monte Carlo simulation is used to simulate γ*, and M interference signals are sent M times to obtain M statistics. Get the observed empirical distribution f of T T (t), given the significance level τ, the decision threshold γ* is obtained by simulation.

[0088]

[0089] It is an estimate of the detection probability, which is used to describe the probability that the radar is judged as a cognitive radar after sending M interference signals.

[0090] The results show that the KS detector has better performance in identifying cognitive radars and non-cognitive radars. When the probability of the second type of error is reduced to 1%, the performance of the KS detector is improved by about 10db compared with the Afriat detector.

[0091] Steps for identifying cognitive radar based on KS test

[0092] In order to optimize the interference signal to reduce the probability of the second type of error in cognitive radar recognition, we used the momentum gradient descent algorithm. The algorithm is given as follows:

[0093] Step 1: Randomly select the initial interference signal

[0094] α k＝1 =[α 1 ,α 2 ,α 3 ,...,α i .....α N ] T (i=1,2,3....N)

[0095] Step 2: Define the objective function

[0096]

[0097] Interference signal α k is the kth interference signal sent by the jammer (k=1,2,3....M) and each interference signal α k is an N-dimensional vector, ,α k =[α 1 ,α 2 ,α 3 ,...,α i .....α N ] T Each element in represents the value of the interference signal power spectrum density at each frequency point. There are N frequency points in total, so there are N spectrum values, which together constitute the interference signal power spectrum density.

[0098] Step 3: Estimate the initial gradient using the finite difference method

[0099]

[0100] In α k＝1 Add a random perturbation Δ to perform finite difference calculation for gradient calculation. Δ randomly takes values ​​of -0.005 and 0.005 with a probability of 0.5.

[0101] Step 4: Parameter update

[0102]

[0103] α k+1 =α k -w·v k+1

[0104] Herev k+1 is the momentum term, which represents the cumulative direction and size of the previous gradient. The momentum factor (a number between 0 and 1, commonly set to 0.9) is used to control the influence of the previous time step gradient on the current update. w represents the learning rate used to control the size of the parameter update.

[0105] Step 5: Update subsequent gradients

[0106]

[0107] The gradient is updated and steps 3 to 5 are repeated until the convergence condition or the set maximum number of iterations is reached.

[0108] Through the momentum gradient descent algorithm, it is possible to optimize the interference detection signal and reduce the probability of the second type of error in radar identification.

[0109] Result Plot

[0110] Power spectral density of cognitive radar signal and jammer signal

[0111] Figure 2The comparative simulation of cognitive radar signal power spectrum and target spectrum and interference signal power spectrum is given. The total energy of radar signal and interference signal is set to E = 5000, and the total energy is the total power carrier signal bandwidth B = 10MHz, Fc = 5MHz, A = 1.21×10 -3 , Interference signal P d (f) = ζ a G a (f)+ζ b G b (f)+ζ c G c (f). G(f) is the noise phase interference, the center frequency is randomly adjusted in the simulation, ζ is the amplitude weighting coefficient, and the value between 0 and 1 is randomly taken in the simulation. For the convenience of representation, the frequency axis is normalized to 1, the signal energy is normalized to [0.1, 1.1], and the environmental noise is considered to be the thermal noise temperature Ts = 300K, P n (f) = 4.1430 × 10 -21 J. The frequency domain sampling points N = 512, and the number of allocations for optimizing power allocation in the MMA algorithm is M = 5120. It can be found that the power spectrum density optimized by cognitive radar is always within the target spectrum P g The higher the power is, the higher the power is allocated. d Less power is allocated where the interference spectrum is strong, and more power is allocated where the interference is weak.

[0112] Radar Identification Based on KS Test

[0113] Given the first type error probability value τ, the correct detection probability of cognitive radar is controlled Radar detection using KS detector That is, the probability curve of non-cognitive radar being correctly identified. The basic radar parameter settings are the same as the previous configuration. The range is between 10 and 30 dB. The interference waveform is sent M = 6000 times under the same σ to calculate the probability value. The sample points L = 100 for each signal transmission. Given the significance level τ, the cognitive radar determines the corresponding threshold γ by using Monte Carlo simulation. In the simulation, the non-cognitive radar uses a pulse-to-pulse modulation frequency agile radar with a carrier frequency of 0.3B to 0.7B randomly selected between pulses. The signal energy is set to the same E = 5000 as the cognitive radar signal energy. Send M interference signals to simulate the probability of non-cognitive radar being correctly identified. Under the same significance level, when the standard deviation of the frequency domain noise is small, the probability of correct identification of cognitive radar is higher, and as the noise increases, the probability of correct detection decreases. This is because when the frequency domain noise is large, the SINR samples obtained are greatly affected by the interference of the noise, resulting in a large deviation between the SINR empirical distribution obtained from the sample and the theoretical SINR distribution, which reduces the probability of correct detection, and the KS test is sensitive to both the number of samples and the sampling deviation. When the noise variance increases, a larger number of samples may be required to effectively reflect the characteristics of the actual distribution, otherwise the difference between the actual distribution and the theoretical distribution may occur due to sampling deviation. If the noise is small, the SINR sample distribution is less disturbed and more consistent with the theoretical SINR distribution, so the probability of correct detection will be improved.

[0114] Radar Identification Based on KS Test

[0115] Figure 4 The momentum gradient descent is used to optimize the interference signal to reduce the second type error probability P{H 0 ∣H 1}. At the significance level τ = 0.05, In the case of sending interference signals, the interference signals for each iteration are M = 100 times, and the estimated probability of the second type of error is: The momentum factor β is set to 0.9, the learning rate is 0.01, and Δ takes random values ​​of -0.005 and 0.005 with a probability of 0.5. The threshold γ = 0.989, and the initial interference signal can be randomly selected. One group performs momentum gradient descent optimization on the initial interference signal, and the other group randomly selects the interference signal. It can be found that the probability of the second type of error obtained by the unoptimized interference signal fluctuates around 0.75. When the momentum gradient descent is iterated to 40 to 100 times, the algorithm roughly converges, and the probability of the second type of error drops to about 30%, which is nearly 50% lower than the error probability of the unoptimized interference signal.

[0116] Performance comparison between KS detector and Afriat theorem detector

[0117] Figure 5 The second type error probability curve P{H is compared between KS detector and Afriat detector under the condition of controlling the same first type error at the significance level τ = 0.05. 0 ∣H 1The principle of the Afriat theorem detector is to use the characteristics of cognitive radar to optimize its own waveform parameters according to the external environment, abstract it into rational consumer behavior in economics that can maximize its own utility, and then use the Afriat theorem inequality to determine whether the radar's behavior has cognitive characteristics. The non-cognitive radar here is set as a frequency agile radar. B = 10MHz is the signal bandwidth. It should be noted that each group of corresponding σ sent 1000 different interference waveforms to identify the radar, through Send M jamming signals, count the number of times it is judged as cognitive radar and divide it by the total number of signal transmissions to simulate the probability of type II error The total energy of non-cognitive radar and cognitive radar is E = 5000. It can be found that From 20 to 40db, the performance of KS detector is always better than that of Afriat detector. The probability of type II error always remains at a low level. However, using Afriat's theorem for hypothesis testing, When the noise variance reaches 30dB, the probability of type II error is always high. In the case of low noise variance, it is expected that cognitive and non-cognitive states can be easily distinguished. The probability of type II error begins to increase rapidly, and cognitive radar and non-cognitive radar become difficult to distinguish. As σ becomes larger, the increase in noise variance increases the test statistic, and more data from non-cognitive radars in group L will be judged to have rational data relationships. That is, it is mistakenly identified as cognitive radar, resulting in an increased probability of type II error.

[0118] Figure 6 Schematic diagram of radar identification based on KS test and momentum gradient descent provided by an embodiment of the present invention. An embodiment of the present invention provides a radar identification device based on KS test and momentum gradient descent, including: an interference filtering module 610, an estimation module 620 and a testing module 630.

[0119] The interference filtering module is used to perform interference filtering on the original echo signal collected by the radar of the UAV to obtain a usable echo signal after filtering; the original echo signal includes: an interference signal spectrum; the usable echo signal includes: a radar signal spectrum;

[0120] The estimation module is used to estimate the relevant parameters of the radar signal and derive the distribution of the signal to interference plus noise ratio SINR according to the available echo signal after filtering, and obtain the radar SINR sample distribution and the pre-derived theoretical signal to interference plus noise ratio SINR distribution;

[0121] The test module is used to sense the radar's adaptive power allocation capability using the KS test based on the radar SINR sample distribution and the pre-derived theoretical signal to interference plus noise ratio SINR distribution, and optimize the interference signal using the momentum gradient descent method.

[0122] An embodiment of the present invention provides an electronic device, comprising: a processor and a memory, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor executes the machine-readable instructions to execute the steps of the radar identification method based on KS test and momentum gradient descent as described above.

[0123] An embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the radar identification method based on KS test and momentum gradient descent are executed as described above.

[0124] Obviously, those skilled in the art can make various changes and modifications to the embodiments of the present invention without departing from the spirit and scope of the embodiments of the present invention. Thus, if these modifications and variations to the embodiments of the present invention fall within the scope of the claims of the embodiments of the present invention and their equivalents, the embodiments of the present invention are also intended to include these modifications and variations.

Claims

1. A radar identification method based on KS test and momentum gradient descent, characterized in that: include: Perform interference filtering on the original echo signal collected by the radar of the UAV to obtain a usable echo signal after filtering; The original echo signal includes: an interference signal spectrum; the available echo signal includes: a radar signal spectrum; According to the available echo signal after filtering, the relevant parameters of the radar signal are estimated and the distribution of the signal to interference plus noise ratio SINR is derived, and the radar SINR sample distribution and the pre-derived theoretical signal to interference plus noise ratio SINR distribution are obtained; According to the radar SINR sample distribution and the pre-derived theoretical signal to interference plus noise ratio SINR distribution, the KS test is used to perceive the radar's adaptive power allocation capability, and the momentum gradient descent method is used to optimize the interference signal.

2. A cognitive radar recognition method based on KS test and momentum gradient descent as claimed in claim 1, characterized in that: The original echo signal collected by the radar of the UAV is subjected to interference filtering, and the filtered available echo signal includes: Performing frequency domain analysis on the original echo signal collected by the radar of the UAV to obtain relevant parameters of the original echo signal; the relevant parameters of the original echo signal include: at least one of noise power density, interference signal spectral density, and target signal spectral density; An adaptive power allocation algorithm is used to optimize the original echo signal according to the noise power density, interference signal spectral density and target signal spectral density.

3. The cognitive radar recognition method based on KS test and momentum gradient descent according to claim 1, characterized in that: The relevant parameters of the original echo signal also include: noise power density, interference signal spectral density, target signal spectral density; The distribution of signal to interference plus noise ratio SINR is derived based on the noise power density, interference signal spectral density and target signal spectral density.

4. The cognitive radar recognition method based on KS test and momentum gradient descent according to claim 1, characterized in that: According to the available echo signal after filtering, the relevant parameters of the radar signal are estimated and the distribution of the signal to interference plus noise ratio SINR is derived, and the radar SINR sample distribution and the pre-derived theoretical signal to interference plus noise ratio SINR distribution are obtained, including: using a KS detector, calculating the test statistic and the threshold under a given significance level τ.

5. The cognitive radar recognition method based on KS test and momentum gradient descent according to claim 1, characterized in that: According to the radar SINR sample distribution and the pre-derived theoretical signal to interference plus noise ratio SINR distribution, the KS test is used to perceive the radar's adaptive power allocation capability, and the momentum gradient descent method is used to optimize the interference signal, including: using the momentum gradient descent algorithm to optimize the jammer's signal to reduce the probability of Type II error in radar identification, and comparing it with a cognitive radar detector based on the economic Afriat theorem.

6. A radar identification device based on KS test and momentum gradient descent, characterized in that: include: The interference filtering module is used to perform interference filtering on the original echo signal collected by the radar of the UAV to obtain a usable echo signal after filtering; The original echo signal includes: an interference signal spectrum; the available echo signal includes: a radar signal spectrum; An estimation module is used to estimate the relevant parameters of the radar signal and derive the distribution of the signal to interference plus noise ratio SINR according to the available echo signal after filtering, and obtain the radar SINR sample distribution and the pre-derived theoretical signal to interference plus noise ratio SINR distribution; The test module is used to sense the adaptive power allocation capability of the radar using the KS test according to the radar SINR sample distribution and the pre-derived theoretical signal to interference plus noise ratio SINR distribution, and optimize the interference signal using the momentum gradient descent method.

7. An electronic device, characterized in that: include: A processor and a memory, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor executes the machine-readable instructions to execute the steps of the radar identification method based on KS test and momentum gradient descent as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the radar identification method based on KS test and momentum gradient descent as claimed in any one of claims 1 to 5 are executed.

Images