Parameter blind estimation method for unmanned aerial vehicle remote control image transmission signal

Abstract

The application provides a parameter blind estimation method for a remote control image transmission signal of a UAV, a time-frequency matrix is obtained by calculating a short-time Fourier transform of a received signal; a denoising process is performed on the time-frequency matrix, a k-means algorithm based on a residence time length or a k-means clustering algorithm based on a time-frequency variance is used to cluster the time-frequency matrix after denoising to remove a fixed-frequency interference, and a reconstructed time-frequency matrix is obtained by reconstruction; a time-frequency ridge of the reconstructed time-frequency matrix is extracted; a Haar wavelet transform is performed on the time-frequency ridge, and a modulus value after the Haar wavelet transform is taken; a time interval between peaks of the modulus value and a time corresponding to the peaks of the modulus value are obtained by using a difference method; a take-off time of the remote control signal is estimated according to the time interval and the time corresponding to the peaks of the modulus value; and an actual frequency of the remote control signal is estimated based on the take-off time. Compared with the prior art, the application can remove the fixed-frequency interference and estimate the parameters of the remote control signal.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of unmanned aerial vehicle, and particularly relates to a parameter blind estimation method for remote control image transmission signals of unmanned aerial vehicles. BACKGROUND

[0002] With the popularity of unmanned aerial vehicles, the premise of handling the illegal behavior of unmanned aerial vehicles is to detect the specific unmanned aerial vehicle that violates the rules. Currently, there are unmanned aerial vehicle detection methods including photoelectric detection, radar detection and radio spectrum detection. In order to improve the applicability of unmanned aerial vehicles, commercial unmanned aerial vehicle companies have been improving the technology in terms of miniaturization and maneuverability.

[0003] The existing unmanned aerial vehicles are becoming smaller and more maneuverable, and the recognition rate will be greatly reduced if the scheme of detecting low, small and slow unmanned aerial vehicles is adopted, such as photoelectric detection or radar detection. The mainstream consumer-level unmanned aerial vehicles on the market will carry a motion camera and a corresponding image signal transmission system, and will always transmit signals to the remote ground station during the flight of the unmanned aerial vehicle. Therefore, the unmanned aerial vehicle can be detected and identified according to the unmanned aerial vehicle image transmission signal, the frequency hopping pattern reflects the important time-frequency characteristics of the unmanned aerial vehicle remote control image transmission signal, and the frequency hopping frequency, hopping speed and frequency hopping period of the unmanned aerial vehicle remote control image transmission signal can be obtained according to the frequency hopping pattern, which provides important information for the defense party to implement countermeasures against the illegal unmanned aerial vehicle. Therefore, the technical personnel can realize the detection of the illegal unmanned aerial vehicle by estimating the parameters of the unmanned aerial vehicle remote control signal.

[0004] In the prior art, the parameter estimation method of the unmanned aerial vehicle remote control signal is as follows: first, the received signal is subjected to short-time Fourier transform (STFT) to transform the received signal into the time-frequency domain, and then the time-frequency ridge line of the time-frequency distribution is extracted. The time-frequency ridge line of the remote control signal appears as a broken line segment that constantly changes on the time-frequency surface, the horizontal direction represents the time dimension, and the vertical direction represents the frequency dimension. The instantaneous frequency of the signal can be obtained by analyzing the time-frequency ridge line, and then the variation law of the signal with time can be obtained, and then the parameters of the remote control signal can be estimated.

[0005] In the prior art, due to the existence of the fixed-frequency interference, the time-frequency ridge line is concentrated around the frequency of the fixed-frequency interference, which destroys the original time-frequency ridge line distribution characteristics. When the interference power is further enhanced, the time-frequency ridge line is approximately a straight line, so that the parameter estimation method based on the time-frequency ridge line is invalid. SUMMARY

[0006] In order to solve the above problems existing in the prior art, the present application provides a parameter blind estimation method for remote control image transmission signals of unmanned aerial vehicles. The technical problem to be solved by the present application is solved by the following technical scheme:

[0007] The application provides a parameter blind estimation method for a remote control image transmission signal of a UAV, which comprises the following steps:

[0008] Step 1: receiving a signal and calculating a short-time Fourier transform of the received signal to obtain a time-frequency matrix;

[0009] Step 2: performing denoising processing on the time-frequency matrix to obtain a denoised time-frequency matrix;

[0010] Step 3: performing clustering on the denoised time-frequency matrix based on a k-means algorithm of a dwell time length or a k-means clustering algorithm based on time-frequency variance to remove fixed-frequency interference and reconstructing a reconstructed time-frequency matrix;

[0011] Step 4: extracting a time-frequency ridge line of the reconstructed time-frequency matrix;

[0012] Step 5: performing Haar wavelet transform on the time-frequency ridge line and taking a modulus value after the Haar wavelet transform;

[0013] Step 6: using a difference method to obtain a time interval between each peak of the modulus value and a time corresponding to each peak of the modulus value;

[0014] Step 7: estimating a take-off time of the remote control signal according to the time interval and the time corresponding to each peak of the modulus value;

[0015] Step 8: estimating an actual frequency of the remote control signal based on the take-off time.

[0016] The application has the following beneficial effects:

[0017] The application provides a parameter blind estimation method for a remote control image transmission signal of a UAV, which comprises the following steps: calculating a short-time Fourier transform of a received signal to obtain a time-frequency matrix; performing denoising processing on the time-frequency matrix, performing clustering on the denoised time-frequency matrix based on a k-means algorithm of a dwell time length or a k-means clustering algorithm based on time-frequency variance to remove fixed-frequency interference and reconstructing a reconstructed time-frequency matrix; extracting a time-frequency ridge line of the reconstructed time-frequency matrix; performing Haar wavelet transform on the time-frequency ridge line and taking a modulus value after the Haar wavelet transform; using a difference method to obtain a time interval between each peak of the modulus value and a time corresponding to each peak of the modulus value; estimating a take-off time of the remote control signal according to the time interval and the time corresponding to each peak of the modulus value; and estimating an actual frequency of the remote control signal based on the take-off time. Compared with the prior art, the application can remove fixed-frequency interference and estimate parameters of the remote control signal.

[0018] The application will be further described in detail below with reference to the drawings and embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1is a schematic diagram of a parameter blind estimation method of a remote control image transmission signal of a UAV provided by the application;

[0020] Figure 2a is a diagram of the distribution of residual noise, remote control signals and constant frequency interference in the time dimension at a specific frequency;

[0021] Figure 2b is a clustering result based on residence time length of the application;

[0022] Figure 2c is a clustering result based on time-frequency variance of the application;

[0023] Figure 3a is a time-frequency diagram and time-frequency ridge line diagram reconstructed by the remote control signal parameter blind estimation method of improving the time-frequency ridge line in the application;

[0024] Figure 3b is a time-frequency diagram and time-frequency ridge line diagram reconstructed by the remote control signal parameter blind estimation method based on time-frequency variance clustering in the application;

[0025] Figure 4a is an estimation result diagram of the hopping period of the application;

[0026] Figure 4b is an estimation result diagram of the frequency hopping frequency of the application;

[0027] Figure 5a is a comparison diagram of the period estimation performance under different interference powers of the application;

[0028] Figure 5b is a comparison diagram of the frequency estimation performance under different interference powers of the application. DETAILED DESCRIPTION

[0029] The application will be described in further detail below with specific embodiments, but the embodiments of the application are not limited thereto.

[0030] The application provides two schemes, which are a remote control signal parameter blind estimation method of improving the time-frequency ridge line and a remote control signal parameter blind estimation method based on time-frequency variance clustering.

[0031] 1. The technical principle of the remote control signal parameter blind estimation method of improving the time-frequency ridge line is: first, after receiving the signal, calculate the STFT of the received signal to obtain the corresponding time-frequency matrix; then, pre-process the time-frequency matrix to obtain the denoised time-frequency matrix; cluster the denoised matrix according to the k-means algorithm based on residence time length, remove the constant frequency interference, reconstruct the time-frequency matrix to obtain the reconstructed time-frequency matrix; then extract the time-frequency ridge line of the reconstructed time-frequency matrix, and perform Haar wavelet transform on the time-frequency ridge line to obtain the modulus |W(a,t)|; use the difference method to find the time interval T between the peaks of |W(a,t)|h , the reciprocal of which is the hop speed of the remote control signal; the time corresponding to each peak of |W(a, t)| is calculated to estimate the take-off time of the remote control signal; the midpoint of each hop cycle is taken to estimate the frequency of the remote control signal.

[0032] 2. The technical principle of the remote control signal parameter blind estimation method based on time-frequency variance clustering is: first, after receiving the signal, the STFT of the received signal is calculated to obtain the corresponding time-frequency matrix; then the optimal threshold is solved by using a genetic algorithm to obtain a new time-frequency matrix; the time-frequency variance S of the time-frequency matrix is calculated, and k-means algorithm is used for clustering (the number of clusters is 2) to remove the fixed-frequency interference and reconstruct the reconstructed time-frequency matrix; the time-frequency ridge of the reconstructed time-frequency matrix is extracted, and Haar wavelet is used for detection; the period and hop speed of the frequency hopping signal are estimated according to the formula; the reconstructed time-frequency matrix is subjected to ISTFT, and the frequency of the remote control signal is obtained by converting to the frequency domain, thereby completing the estimation of the parameters of the remote control signal.

[0033] As shown in Figure 1 , the present application provides a kind of unmanned aerial vehicle remote control image transmission signal parameter blind estimation method, comprising:

[0034] Step 1, receiving signal, and the short-time Fourier transform of the received signal is calculated to obtain a time-frequency matrix;

[0035] Step 2, the time-frequency matrix is denoised to obtain a denoised time-frequency matrix;

[0036] Step 3, the denoised time-frequency matrix is clustered based on the k-means algorithm of the residence time length or the k-means clustering algorithm based on the time-frequency variance to remove the fixed-frequency interference, and a reconstructed time-frequency matrix is reconstructed;

[0037] Step 4, extracting the time-frequency ridge of the reconstructed time-frequency matrix;

[0038] Step 5, Haar wavelet transform is performed on the time-frequency ridge, and the modulus value after Haar wavelet transform is taken;

[0039] Step 6, the time interval between each peak of the modulus value and the time corresponding to each peak of the modulus value are calculated by the difference method;

[0040] Step 7, the take-off time of the remote control signal is estimated according to the time interval and the time corresponding to each peak of the modulus value;

[0041] Step 8, the actual frequency of the remote control signal is estimated based on the take-off time.

[0042] The differences and detailed processes of the two schemes are described in detail below.

[0043] (1) The remote control signal parameter blind estimation method for improving the time-frequency ridge line comprises the following steps:

[0044] S1, calculating the STFT of the signal to obtain the corresponding time-frequency matrix STFT x (t,f);

[0045] S2, carrying out the denoising processing of the time-frequency matrix to obtain the denoised time-frequency matrix STFT

[0046] In the case of high signal-to-noise ratio, the genetic algorithm is used to remove the noise elements in the time-frequency matrix; in the case of low signal-to-noise ratio, the time-frequency matrix denoising method based on iteration is used to remove the noise elements in the time-frequency matrix.

[0047] The genetic algorithm can be used to solve the optimal truncation threshold ε, and the time-frequency matrix STFT x (t,f) is denoised by using the truncation threshold ε to obtain the denoised time-frequency matrix STFT(t,f).

[0048] The genetic algorithm is used to solve the optimal truncation threshold ε, and the time-frequency matrix STFT x (t,f) is denoised by using the truncation threshold ε to obtain the denoised time-frequency matrix STFT(t,f), which comprises:

[0049] Step 2.1, taking the maximum value M and the minimum value m of the time-frequency matrix STFT x (t,f), and calculating the initial threshold value ε1 of the truncation threshold ε according to the formula ε1=(M+m) / 2;

[0050] Step 2.2, dividing the time-frequency matrix STFT x (t,f) into two parts X s and X n with the initial value ε1 as the boundary, X s represents the time-frequency component of the signal containing the fixed-frequency interference, and X n represents the time-frequency component of the noise;

[0051] Step 2.3, counting the number N s and N n of the two parts X s and X n , and then calculating the average values of the two parts X s and X n , which are represented as X s1 and X n1 respectively.

[0052] Step 2.4, taking the two parts X s1 and X n1 as new signal components to obtain X s1 and X n1The average value of the values ​​is used to obtain a new threshold ε2, and the initial threshold ε1 is replaced with the new threshold ε2.

[0053] Step 2.5: Repeat steps 2.2 to 2.5 to continuously update and obtain a new threshold until ε is reached. k+1 =ε k When the iteration process ends, the final threshold ε = ε is obtained. k+1 ;

[0054] Step 2.6: The final threshold is used as the truncation threshold ε. Elements in the time-frequency matrix that are below the truncation threshold ε are removed to obtain the denoised time-frequency matrix STFT(t,f).

[0055] S3. The denoised time-frequency matrix STFT(t,f) is clustered using the k-means algorithm based on dwell time to remove fixed-frequency interference, thereby reconstructing the time-frequency matrix and obtaining the reconstructed time-frequency matrix STFT. R (t,f);

[0056] like Figure 2a As shown, the dwell time of a signal in the time-frequency graph characterizes its dwell time. Noise is discretely distributed in the time dimension, resulting in a very short dwell time. The remote control signal is a frequency-hopping signal, existing in segments on the time-frequency surface, with a dwell time equal to one frequency-hopping cycle. Fixed-frequency interference exists throughout the entire observation time, forming a long line segment across the time axis. Therefore, we can first statistically analyze the dwell time of each frequency component in the time-frequency matrix. For an M×N time-frequency matrix, we first determine if an element is 0. If it is 0, we proceed to the next element; if not, we record the dwell time as 1. We continue to check the next element, returning the dwell time as 1 if it is 0, and incrementing it by 1 if it is not 0, until the dwell time is output. Based on the different dwell times, we use the k-means clustering algorithm to separate noise, remote control signals, and fixed-frequency interference.

[0057] The specific process of using a clustering algorithm based on dwell time to remove fixed-frequency interference and reconstruct the reconstructed time-frequency matrix is ​​as follows:

[0058] Step 3.1a: Calculate the dwell time of each frequency component in the denoised time-frequency matrix STFT(t,f), denoted as {len1, len2, len3, ..., len...} n}, randomly select 3 center points {μ1,μ2,μ3} to divide the data into 3 clusters {λ1,λ2,λ3};

[0059] Step 3.2a: Calculate the Euclidean distance d between all data points and the cluster center. ij =|len i -μ j|, and len i The samples in each cluster λ i are classified into corresponding categories λ

[0060] Step 3.3a, the average value of all sample points in each cluster partition λ i is calculated as a new center point, and the process of steps 3.1a and 3.2a is repeated until all center points no longer change, and the clustering is completed, as shown in Figure 2b .

[0061] Step 3.4a, the clusters are reserved as frequency components of remote control signals, and the clusters are set to 0 as frequency components of constant frequency interference and residual noise, and the original time-frequency matrix is updated to obtain a reconstructed time-frequency matrix STFT R (t,f), as shown in Figure 3a .

[0062] S4, extracting the time-frequency ridge line of the reconstructed time-frequency matrix , as shown in Figure 3a .

[0063] For STFT, its time-frequency ridge line refers to the peak frequency at each time in its time-frequency distribution, that is, X(t,f) represents the time-frequency distribution of the signal.

[0064] Step 5, Haar wavelet transform is performed on f i (t), and the modulus value is taken;

[0065] There is a jump between adjacent frequency components, which produces a certain step phenomenon similar to a square wave. Since wavelet transform has a function similar to a "magnifying glass" and can amplify the details of the signal to a certain extent, selecting an appropriate wavelet function can detect some singular points in the signal. There are many wavelet functions, and different wavelet functions can be selected for different signal characteristics. Since the time-frequency ridge line is approximately a square wave, and Haar wavelet has good detection capability for such jumps, it can detect the time points of the jumps locally, so Haar wavelet can be used to detect the time points of the jumps and accurately estimate the parameters such as the frequency hopping period of the signal. The formula of wavelet transform is

[0066]

[0067] In the formula, a represents the scale parameter (usually a = 1), f i (t) represents the extracted time-frequency ridge line, * represents conjugate, τ represents displacement, and ψ(t) represents the mother function of Haar wavelet, whose expression is

[0068]

[0069] If in the same cycle, the frequency of the remote control signal does not change, and the frequency is f k , then the corresponding wavelet transform result is:

[0070]

[0071] If in different cycles, the frequency of the remote control signal changes because the integral interval spans different hop cycles, and suppose at τ0, the frequency changes from f k to f k+1 .

[0072] When , the wavelet transform result is:

[0073]

[0074] When , the wavelet transform result is:

[0075]

[0076] From the above calculation, in the same cycle, |W(a, τ)| is always 0, while in different hop cycles, the value of |W(a, τ)| is not 0, and reaches the maximum value when at the position where the frequency jumps. With the sliding of τ in the entire interval, |W(a, τ)| will constantly change between 0 and the peak value. The change of |W(a, τ)| reflects the change of the frequency of the remote control signal, because in the same hop frequency cycle, the frequency of the remote control signal does not change, and only in different hop frequency cycles, the frequency of the remote control signal is different, so the interval between the two mutation moments is also the hop frequency cycle of the signal.

[0077] S6, use the difference method to find the time interval T between each peak of |W(a, t)| h , and the reciprocal is the hop speed of the remote control signal;

[0078] The time interval of each spectrum peak is the estimated hop cycle T h of the hop frequency signal, and the reciprocal is the hop speed. In order to obtain a more accurate cycle estimation value, the difference method is used to calculate the odd and even terms, that is,

[0079]

[0080] Wherein, n represents the number of peak values, t pp (k) represents the time when the kth peak value appears.

[0081] Step 7, find the time corresponding to each peak of |W(a, t)| to estimate the take-off time of the remote control signal;

[0082] Ignoring the time delay of the receiver, the time corresponding to the first peak (the second frequency hopping) is the first hop (start) time of the remote control signal. Due to the existence of noise, the extraction of the time-frequency ridge line will have a small error, so the time of each peak is not a completely accurate estimate. The time can be averaged according to multiple peaks to offset the error caused by noise, that is

[0083]

[0084] S8, taking the midpoint of each hop period, estimating the frequency of the remote control signal.

[0085] For the estimation of the frequency of the remote control signal, the time-frequency ridge line extracted can be directly estimated, and considering the influence of noise, the frequency corresponding to the middle point of each hop period is usually taken as the estimated value of the actual frequency, that is

[0086]

[0087] (2) The remote control signal parameter blind estimation method based on time-frequency variance clustering includes the following steps:

[0088] S1, calculating the STFT of the received signal to obtain the corresponding time-frequency matrix STFT x (t,f).

[0089] S2, solving the optimal truncation threshold ε by using a genetic algorithm to obtain the time-frequency matrix STFT(t,f).

[0090] The present application can solve the optimal truncation threshold ε by using a genetic algorithm, and denoise the time-frequency matrix STFT x (t,f) by using the truncation threshold ε to obtain the time-frequency matrix STFT(t,f) after denoising.

[0091] Solving the optimal truncation threshold ε by using a genetic algorithm, and denoising the time-frequency matrix STFT x (t,f) by using the truncation threshold ε to obtain the time-frequency matrix STFT(t,f) after denoising includes:

[0092] Step 2.1, taking the maximum value M and the minimum value m of the time-frequency matrix STFT x (t,f), the initial threshold value ε1 of the truncation threshold ε is calculated according to the formula ε1=(M+m) / 2;

[0093] Step 2.2, divide the time-frequency matrix STFT x (t,f) into two parts X s and X n , X s represents the time-frequency component of the signal containing the fixed-frequency interference, and X n represents the time-frequency component of the noise;

[0094] Step 2.3, statistics X s and X n The number of parts N s and N n Then calculate X s and X n The average of these two parts, respectively, represented as X s1 and X n1 ;

[0095] Step 2.4, X s1 and X n1 These two parts as a new signal component, X s1 and X n1 The average value is obtained, thereby obtaining a new threshold ε2, using the new threshold ε2 to replace the initial threshold ε1;

[0096] Step 2.5, repeat steps 2.2 to 2.5 to constantly update the new threshold until ε k+1 = k , end the iteration process, and obtain the final threshold ε = ε k+1 ;

[0097] Step 2.6, obtain the final threshold as a truncation threshold ε, remove the elements in the time-frequency matrix that are lower than the truncation threshold ε, and obtain the denoised time-frequency matrix STFT(,f).

[0098] S3, calculate the time-frequency variance S of STFT(,f), cluster S by k-means algorithm (the number of clusters is 2), remove the fixed frequency interference and reconstruct to obtain the matrix STFT R (,f).

[0099] Assume that the remote control signal after denoising is as follows:

[0100]

[0101] Define the time-frequency variance of the time-frequency matrix STFT(,f) as S, then

[0102]

[0103] Where j = 1, 2, …, n. Further, the expression of all time-frequency variances can be obtained:

[0104]

[0105] Where J(,f) represents the interference signal.

[0106] The time-frequency variance represents the dispersion degree of the signal along the time dimension at a certain frequency. For the frequency hopping signal, its distribution is concentrated in a fixed time-frequency domain, and the time-frequency variance is large. For the fixed frequency interference and noise, since they are uniformly distributed in the whole time dimension, the time-frequency variance is small. By means of the k-means clustering algorithm, the two can be well distinguished.

[0107] The k-means clustering algorithm based on the time-frequency variance clusters the time-frequency matrix after denoising to remove the fixed frequency interference, and the reconstructed time-frequency matrix is obtained. The specific process is as follows:

[0108] Step 3.1b, calculate the time-frequency variance s of each row in the time-frequency matrix STFT(t,f) after denoising i (i∈[1,N]) and the set of time-frequency variances is denoted as {s1,s2,s3....s N} Randomly select 2 center points {α1,α2} to divide the data into 2 clusters {β1,β2};

[0109] Step 3.2b, calculate the Euclidean distance D of all data points from the clustering center points ij =|s i -α j |,(j∈1,2) and divide s i into the corresponding category β j ;

[0110] Step 3.3b, calculate the average value of all sample points in each cluster partition β j as the new center point;

[0111] Step 3.4b, repeat the process of steps 3.2b and 3.3b until all center points no longer change, and the clustering is completed, as shown in Figure 2c .

[0112] Step 3.5b, retain the frequency components clustered as remote control signals, set the frequency components clustered as fixed frequency interference and residual noise to 0, and update the original time-frequency matrix to obtain the reconstructed time-frequency matrix STFT R (t,f), as shown in Figure 3b .

[0113] S4, extract the time-frequency ridge line , as shown in Figure 3b .

[0114] For STFT, the time-frequency ridge line refers to the peak frequency of each time in its time-frequency distribution, that is , where X(t,f) represents the time-frequency distribution of the signal.

[0115] S5, for f i(t) Haar wavelet transform is made and its modulus |W(a, t)| is taken;

[0116] There is a jump between adjacent frequency components, thus a certain step phenomenon similar to square wave is generated. Since wavelet transform has a function similar to "magnifying glass", it can amplify the signal detail part to a certain extent, thus selecting a suitable wavelet function can detect some singular points in the signal. There are many wavelet functions, different wavelet functions can be selected for different signal characteristics. Since the time-frequency ridge is approximately a square wave, Haar wavelet has good detection ability for such jumps, and can detect the time point of the jump locally, thus Haar wavelet can be used to detect the jump time and accurately estimate the frequency hopping period and other parameters of the signal. The formula of wavelet transform is

[0117]

[0118] In the formula, a represents a scale parameter (usually a = 1), f i represents the extracted time-frequency ridge, * represents conjugate, τ represents displacement, and ψ(t) represents the mother function of Haar wavelet, whose expression is:

[0119]

[0120] If in the same period, the frequency of the remote control signal does not change, and the frequency at this time is f k , then the corresponding wavelet transform result is:

[0121]

[0122] If in different periods, the frequency of the remote control signal changes due to the integral interval crossing different hopping periods, and it is assumed that at τ0, the frequency changes from f k to f k+1 .

[0123] When , the wavelet transform result is:

[0124]

[0125] When , the wavelet transform result is:

[0126]

[0127] From the above calculation, in the same cycle, |W(a, τ)| is always 0, and in different hop cycles, the value of |W(a, τ)| is not 0, and when being at the position where the frequency jumps, the maximum value is obtained. With the sliding of τ in the whole interval, |W(a, τ)| will change between 0 and the peak value. The change of |W(a, τ)| reflects the change of the frequency of the remote control signal, because in the same frequency hopping cycle, the frequency of the remote control signal does not change, and only in different frequency hopping cycles, the frequency of the remote control signal is different, so the interval of the two mutation moments is the frequency hopping cycle of the signal.

[0128] S6, the time interval between each peak of |W(a, t)| is T h , and the reciprocal is the hop speed of the remote control signal;

[0129] The time interval of each spectrum peak is the estimated T of the frequency hopping signal h , and the reciprocal is the hop speed. In order to obtain a more accurate cycle estimation value, the difference method of odd and even item cancellation is used, that is,

[0130]

[0131] Wherein, represents the number of peak values, t pp () represents the time when the kth peak value appears.

[0132] S7, the time corresponding to each peak of |W(a, t)| is estimated to estimate the take-off time of the remote control signal;

[0133] Ignoring the time delay of the receiver, the time corresponding to the first peak (the second frequency hopping) is the first hop (take-off) time of the remote control signal. Due to the existence of noise, there will be a small error in the extraction of the time-frequency ridge line, so the time when each peak value appears is not a completely accurate estimate. The time average can be taken according to multiple peak values to offset the error caused by noise, that is,

[0134]

[0135] S8, intercepts a certain hop to reconstruct the time-frequency matrix STFT R , and performs ISTFT to obtain the frequency of the remote control signal, and completes the estimation of the parameters of the remote control signal.

[0136] The formula of ISTFT is as follows,

[0137]

[0138] According to the time interval T hThe frequency hopping period is determined. The reconstructed time-frequency matrix of a certain hop is intercepted, and the inverse short time Fourier transform is performed on the reconstructed time-frequency matrix of the certain hop to obtain the actual frequency of the remote control signal.

[0139] On the basis of having determined the frequency hopping period, the reconstructed time-frequency matrix of a certain hop signal is intercepted, and then the inverse short time Fourier transform (ISTFT) is performed on the reconstructed time-frequency matrix to convert it to the frequency domain, and the frequency value is obtained through the fast Fourier transform. Because the resolution of the fast Fourier transform can be adjusted according to the requirement, the estimation value of the frequency obtained by using this method is more accurate than the frequency estimation by using the time-frequency ridge.

[0140] The effect of the application is further illustrated by simulation experiments.

[0141] 1. Simulation conditions

[0142] The experimental data are the intermediate frequency signals after frequency down-conversion. The normalized frequency set of the remote control signal is [0.025, 0.05, 0.075, 0.10, 0.125, 0.15, 0.175, 0.20], the hopping period is 5 ms, the normalized frequency of the constant frequency interference is 0.25, the number of sampling points is 40000, and the STFT uses a Hamming window with a length of 507. The signal-to-noise ratio is defined as the logarithmic form of the ratio of the signal power to the noise power, the interference-to-signal ratio (ISR) is defined as the logarithmic form of the ratio of the interference power to the signal power, and the relative error of the period estimation is defined as The relative error of the frequency estimation is The normalized mean square error (NRMSE) of the estimation result is defined as follows,

[0143]

[0144] wherein, represents the i-th estimation result of the parameter, N s is the simulation number.

[0145] 2. Simulation content

[0146] In experiment 1, the power of the constant frequency interference is set to be the same as the power of the signal. In the signal-to-noise ratio range of [-15, 5] dB, 100 Monte Carlo tests are performed on each group of data by using three algorithms respectively to obtain the normalized mean square error of the hopping period and the normalized mean square error of the frequency hopping frequency. The simulation results are as follows, Figure 4a and Figure 4bThe reference method 1 represents a parameter estimation method, the reference method 2 represents a second difference method, the method 1 of the application represents an improved parameter estimation method of time-frequency ridge, and the method 2 of the application represents a parameter estimation method based on time-frequency variance clustering.

[0147] In experiment 2, the influence of different fixed frequency interference power on the estimation result is considered, and the estimation relative error E1 less than 1% is defined as a correct estimation. In the signal-to-noise ratio range of [-15, 5] dB, the estimation condition of the interference-to-signal ratio in the range of [0, 10] dB is simulated. The correct estimation probability under each ISR condition is obtained by using method 3 on each group of data for 100 times of Monte Carlo test, and the results are shown in Figure 5a and Figure 5b .

[0148] 1. Analysis of simulation results

[0149] As shown in Figure 4a , with the increase of signal-to-noise ratio, the NEMSE of the reference method 1 for period estimation gradually decreases, but the overall NEMSE of the method 1 is relatively large, which shows that when the fixed frequency interference exists, the reference method 1 cannot correctly estimate the period; the reference method 2 can estimate the period, but it is still sensitive to noise due to the many second difference burrs, the NEMSE of the method 1 and the method 2 of the application gradually decreases with the increase of signal-to-noise ratio, and the two algorithms have good noise suppression effect, and still have high estimation accuracy when the signal-to-noise ratio is low, when the signal-to-noise ratio is greater than -5 dB, the NRMSE is less than 0.1, when the signal-to-noise ratio is greater than 0 dB, the NRMSE converges to 0 rapidly, when the signal-to-noise ratio is less than -8 dB, due to the introduction of too much noise, it is difficult to extract the effective time-frequency interval, resulting in the deterioration of the results obtained by the subsequent clustering method based on the dwell time or the time-frequency variance, thereby the NRMSE becomes large, in general, the accuracy of the method 2 of the application is slightly higher than that of the method 1 of the application, and the convergence speed is slightly faster, but the operation amount is large, and the clustering process based on the dwell time needs more time to complete.

[0150] As shown in Figure 4b , the estimation result of the frequency of the remote control signal is similar to the estimation result of the period. The NRMSE of the reference method 1, the reference method 2 and the method 1 of the application finally converges to 0.1. This is because the STFT cannot make the time domain and frequency domain resolution reach the maximum at the same time, and the product of the time domain resolution and the frequency domain resolution is always greater than a non-zero number. Therefore, the NRMSE of the frequency estimation value obtained by using the above method has a non-zero lower limit. The method 2 of the application selects to estimate the frequency hopping frequency in the frequency domain, and there is no problem of insufficient resolution, and when the signal-to-noise ratio is greater than -8 dB, the NRMSE can converge to 0 rapidly.

[0151] Through analysis of Figure 4a and Figure 4b It can be seen that Method 3 outperforms Method 2 in terms of computational load, processing time, and estimation error of frequency hopping. Therefore, in Experiment 2, this invention focuses on analyzing the anti-interference performance of Method 3.

[0152] Depend on Figure 5a and Figure 5b It can be seen that as the interference power increases, a higher signal-to-noise ratio (SNR) is required to obtain a correct estimation result for the hopping cycle, and the probability curve for the correct estimation of the hopping cycle shifts to the right. When the ISR is greater than 6dB, due to the excessive interference power, the genetic algorithm cannot correctly find a suitable threshold and cannot extract an effective time-frequency range, leading to algorithm deterioration. When the ISR is equal to 0dB, the algorithm converges quickly when the SNR is higher than -5dB. When the ISR is less than 6dB, the algorithm can achieve a correct estimation probability of over 90% when the SNR is lower than 5dB. The estimation results for the frequency hopping frequency are similar to the above results. When the interference power is too high, the weak signal is treated as noise and removed, causing distortion and loss of the time-domain signal obtained by ISTFT, resulting in an error in the estimation of the frequency hopping frequency.

[0153] In summary, the parameter estimation algorithm based on time-frequency variance clustering can simultaneously eliminate the influence of noise and interference on the estimation results when the ISR is less than 5dB, and has high estimation accuracy and strong anti-interference ability.

[0154] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0155] Although this application has been described herein in conjunction with various embodiments, those skilled in the art will understand and implement other variations of the disclosed embodiments by reviewing the accompanying drawings, the disclosure, and the appended claims in carrying out the claimed application. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude a plurality.

[0156] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A method for parameter blind estimation of a UAV remote control image transmission signal, characterized in that, Comprising: Step 1, receiving a signal, and calculating a short-time Fourier transform of the received signal to obtain a time-frequency matrix; Step 2, denoising the time-frequency matrix to obtain a denoised time-frequency matrix; Step 3, clustering the denoised time-frequency matrix based on a k-means algorithm based on a residence time or a k-means clustering algorithm based on time-frequency variance to remove fixed-frequency interference, and reconstructing to obtain a reconstructed time-frequency matrix; Step 4, extracting a time-frequency ridge line of the reconstructed time-frequency matrix; Step 5, Haar wavelet transform is performed on the time-frequency ridge line, and the modulus value after the Haar wavelet transform is taken; Step 6, the time interval between each peak of the modulus value and the time corresponding to each peak of the modulus value are obtained by using the difference method; Step 7, estimating the take-off time of the remote control signal according to the time interval and the time corresponding to each peak of the modulus value; Step 8, estimating the actual frequency of the remote control signal based on the take-off time; Step 2 comprises: Solving the optimal truncation threshold by using a genetic algorithm and using the truncation threshold to denoise the time-frequency matrix to obtain a denoised time-frequency matrix ; The optimal truncation threshold is solved by using a genetic algorithm And a truncation threshold is used The time-frequency matrix is denoised to obtain a denoised time-frequency matrix The time-frequency matrix is denoised to obtain a denoised time-frequency matrix The optimal truncation threshold is solved by using a genetic algorithm Step 2.1, Obtain the time-frequency matrix maximum value and minimum value According to the formula Find the cutoff threshold initial threshold ; Step 2.2, with initial value The time-frequency matrix is divided into two parts represents the time-frequency components of the signal containing the constant frequency interference, represents the time-frequency components of the noise;​​​ Step 2.3, Statistics and Number of parts and Then calculate and The average of these two parts, respectively, are denoted as and ; Step 2.4, and These two parts are used as new signal components. and The average value is used to obtain a new threshold. Using a new threshold Replace the initial threshold ; Step 2.

5. Repeat steps 2.2 to 2.5 to continuously update the new threshold until the iteration process is ended and the final threshold is obtained ; Step 2.

6. The final threshold is obtained as the cutoff threshold The elements in the time-frequency matrix that are lower than the cutoff threshold are removed to obtain the denoised time-frequency matrix .

2. The method of claim 1, wherein, In step 3, the k-means algorithm based on the residence time is used to cluster the denoised time-frequency matrix to remove fixed-frequency interference, and a reconstructed time-frequency matrix is obtained by reconstruction, comprising: Step 3.1a, statistical denoising of the time-frequency matrix The residence time of each frequency component is set as Randomly select 3 center points Divide the data into 3 clusters ; Step 3.2a, compute Euclidean distance of all data points to cluster center point and divide into corresponding classes ; Step 3.3a, calculate each cluster partition Repeat the process of step 3.1a, 3.2a until all the center points do not change anymore, complete the clustering. Step 3.4a, reserve the frequency components clustered as remote control signals, set the frequency components clustered as constant frequency interference and residual noise to 0, update the original time-frequency matrix to get the reconstructed time-frequency matrix .

3. The method of claim 1, wherein, In step 3, the k-means clustering algorithm based on time-frequency variance is used to cluster the denoised time-frequency matrix to remove fixed-frequency interference, and a reconstructed time-frequency matrix is obtained by reconstruction, comprising: Step 3.1b, compute the time-frequency matrix after denoising The time-frequency variance of each row in the matrix The set of time-frequency variances is denoted as Randomly select 2 center points Divide the data into 2 clusters ; Step 3.2b, compute the Euclidean distance of all data points to the cluster center point and put into the corresponding class ; Step 3.3b, compute the average of all sample points in each cluster partition the average of all sample points in the middle of the cluster as the new center point; Step 3.4b, repeating the processes of steps 3.2b and 3.3b until all center points no longer change, and completing clustering; Step 3.5b, reserve the frequency components clustered as remote control signals, set the frequency components clustered as constant frequency interference and residual noise to 0, update the original time-frequency matrix to get the reconstructed time-frequency matrix .

4. The method of claim 3, wherein, In step 3.1b, the denoised time-frequency matrix is as follows: ； Time-frequency matrix The time-frequency variance of Then, ； wherein The expression for the total time-frequency variance is: ； wherein represents an interfering signal.

5. The method of claim 2 or 4, wherein, Step 4 comprises: extracting to a reconstructed time-frequency matrix time-frequency ridges of the reconstructed time-frequency matrix ; For STFT, its ridge line in time-frequency is the peak frequency of each time in its time-frequency distribution, that is , represents the time-frequency distribution of the signal.

6. The method of claim 5, wherein, Step 6 comprises: The difference method is used to find The time interval between each peak is ; Seeking The reciprocal of the value is used to obtain the jumping speed of the remote control signal.

7. The method of claim 6, wherein, Step 8 comprises: When the frequency corresponding to the midpoint of each jump cycle is taken as the estimated value of the actual frequency.

8. The parameter blind estimation method of the unmanned aerial vehicle remote control image transmission signal according to claim 6, characterized in that, According to the time interval determining a frequency hopping period; The reconstructed time-frequency matrix of a certain jump is intercepted, and the short-time inverse Fourier transform of the reconstructed time-frequency matrix of the certain jump is obtained to obtain the actual frequency of the remote control signal.

Images