A frequency hopping signal processing method and system based on binary classification

Abstract

The present invention relates to the field of signal processing technology, and discloses a method for processing a frequency hopping signal based on binary clustering, which uses an adaptive threshold binary clustering algorithm to separate the signal data and noise data of the frequency hopping signal, thereby performing denoising on the frequency hopping signal. The present invention solves the problems of the prior art that the instantaneous bandwidth of the frequency hopping signal cannot be estimated and the parameter estimation accuracy is low.

Description

Technical Field

[0001] The present invention relates to the technical field of signal processing, and in particular to a method and system for processing frequency hopping signals based on binary classification. Background Art

[0002] Whether it is the design of blocking interference in the jamming technology of frequency hopping communication, the design of targeting interference in the jamming technology of frequency hopping communication, or the actual scenario of jamming and countermeasures of enemy reconnaissance drones equipped with frequency hopping communication units, if part (or even all) of the frequency hopping parameters of the enemy frequency hopping signal can be obtained before the jamming, such as hopping time, hopping speed, hopping cycle, and the dwell time of each hop, or even the frequency hopping pattern or hopping law in the time-frequency domain, then our jammer can perform targeted jamming based on these prior knowledge of the frequency hopping parameters of the enemy frequency hopping signal, which can greatly improve the jamming efficiency and jamming success rate of our jammer. Therefore, it can be seen that parameter estimation is important in the process of jamming frequency hopping signals.

[0003] The frequency hopping signal of frequency hopping communication has non-stationary random characteristics, and its frequency is rapidly converted according to the pseudo-random sequence code. In order to analyze the detailed characteristics of non-stationary signals, the time-frequency analysis method can be used. Therefore, for the problem of frequency hopping signal parameter estimation, the spectrum information of the frequency hopping signal at a certain moment can be analyzed by the time-frequency analysis method. Traditional time-frequency analysis methods include pseudo Wigner-Ville Distribution (PWVD), smooth pseudo Wigner-Ville distribution (SPWVD), short-time Fourier transform (STFT), Wigner-Ville Distribution (WVD), wavelet transform, etc. In recent years, some signal detection and perception methods have been widely used in the field of Internet of Things and communication.

[0004] In recent years, members of the intelligent sensing signal research team of the Internet of things and intelligent sensor laboratory (IOTS Lab) have made some progress in the field of frequency hopping signal processing. Under the distributed MWC sampling and recovery framework, the laboratory research team members estimated the frequency hopping parameters such as the hopping time, hopping speed, and hopping period of the frequency hopping signal through the time-frequency ridgeline diagram method. This sparse reconstruction method is currently the most mainstream method for estimating frequency hopping signal parameters. However, this mainstream method for estimating frequency hopping signal parameters cannot estimate the instantaneous bandwidth of the frequency hopping signal. Summary of the invention

[0005] In order to overcome the deficiencies of the prior art, the present invention provides a method and system for processing a frequency hopping signal based on binary classification, which solves the problems existing in the prior art such as the inability to estimate the instantaneous bandwidth of the frequency hopping signal and the low parameter estimation accuracy.

[0006] The technical solution adopted by the present invention to solve the above problems is:

[0007] A frequency hopping signal processing method based on binary clustering uses an adaptive fixed threshold binary clustering algorithm to separate the signal data and noise data of the frequency hopping signal, thereby denoising the frequency hopping signal.

[0008] As a preferred technical solution, the following steps are included:

[0009] S1, complex domain representation: the frequency hopping signal is represented as discrete points in the complex domain;

[0010] S2, separation: separating the signal data and noise data of the frequency hopping signal;

[0011] S3, modulus representation: the discrete points in the complex domain of the signal data and the noise data are expressed on the coordinate axis after the modulus is calculated;

[0012] S4, signal data extraction: extracting the signal data, thereby achieving denoising of the frequency hopping signal.

[0013] As a preferred technical solution, in step S2, the signal data and noise data of the frequency hopping signal are separated by a centroid iteration method, and the clustering centers of the frequency hopping signal and the noise signal are obtained respectively.

[0014] As a preferred technical solution, in step S3, the modulus value of each point of the sampling point data in the complex domain is calculated to obtain the grayscale value in the image of each sampling point data, and then all the grayscale values ​​are marked on the coordinate axis.

[0015] As a preferred technical solution, in step S4, the signal data is extracted by multiplying the corresponding grayscale value of the signal data point on the coordinate axis by 1 or 0.

[0016] As a preferred technical solution, the method further comprises the following steps:

[0017] S5, parameter estimation: using the denoised frequency hopping signal data for frequency hopping signal parameter estimation.

[0018] As a preferred technical solution, in step S5, the normalized coordinates (y1, x1, y2, x2) of each frequency hopping segment are first obtained, and then the center frequency and instantaneous bandwidth of the frequency hopping signal are obtained in proportion according to the normalized coordinates, and the dwell time of each hop is obtained, as shown in the following formula:

[0019] f=[(y2+y1) / 2]f s ,

[0020] B=(y2-y1)*f s ,

[0021] R=1 / [(x2-x1)*T s ];

[0022] Among them, f is the center frequency of the frequency hopping signal obtained by proportion according to the normalized coordinates, y1 is the lower boundary of each frequency hopping band, y2 is the upper boundary of each frequency hopping band, and f s is the sampling frequency of the frequency hopping signal, B is the instantaneous bandwidth of the frequency hopping signal obtained by proportionality based on the normalized coordinates, x1 is the left boundary of each frequency hopping segment, x2 is the right boundary of each frequency hopping segment, R is the dwell time of each hop, T s =1 / f s .

[0023] The frequency hopping signal denoising system based on adaptive fixed threshold binary classification is used to implement the frequency hopping signal processing method based on binary classification, and includes the following modules connected in sequence:

[0024] Complex domain representation module: used to represent the frequency hopping signal in the complex domain as discrete points;

[0025] Separation module: used to separate the signal data and noise data of the frequency hopping signal;

[0026] Modulus value representation module: used to calculate the modulus value of discrete points in the complex domain of signal data and noise data and represent them on the coordinate axis;

[0027] Signal data extraction module: used to extract signal data, so as to realize denoising processing of frequency hopping signal.

[0028] As a preferred technical solution, it also includes a parameter estimation module connected to the signal data extraction module, and the parameter estimation module is used to use the denoised frequency hopping signal data for frequency hopping signal parameter estimation.

[0029] Compared with the prior art, the present invention has the following beneficial effects:

[0030] The present invention proposes a corresponding adaptive threshold binary clustering algorithm for the problem of removing noise in frequency hopping signal parameter estimation based on image processing, which greatly improves the parameter estimation accuracy of the frequency hopping signal in a noisy environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 A schematic diagram of the steps of a frequency hopping signal processing method based on binary classification clustering according to the present invention;

[0032] Figure 2 It is one of the iterative process diagrams of the adaptive threshold binary clustering algorithm in the complex domain;

[0033] Figure 3 The second diagram is the iterative process of the adaptive threshold binary clustering algorithm in the complex domain;

[0034] Figure 4 Figure 3 of the iterative process of the adaptive threshold binary clustering algorithm in the complex domain

[0035] Figure 5 The fourth diagram is the iterative process of the adaptive threshold binary clustering algorithm in the complex domain;

[0036] Figure 6 Figure 5 is the iterative process of the adaptive threshold binary clustering algorithm in the complex domain;

[0037] Figure 7 Figure 6 is the iterative process of the adaptive threshold binary clustering algorithm in the complex domain;

[0038] Figure 8 Figure 7 is the iterative process of the adaptive threshold binary clustering algorithm in the complex domain;

[0039] Fig. 9 This is the effect diagram of the adaptive threshold binary clustering algorithm in the complex domain;

[0040] Fig.10 This is the effect diagram of the adaptive threshold binary clustering algorithm;

[0041] Fig.11 This is a comparison of the frequency-frequency diagram of the frequency-hopping signal of multiple network stations before and after denoising;

[0042] Fig.12 It is a schematic diagram of parameter estimation of the frequency hopping signal time-frequency diagram of one of the dual-network stations after denoising;

[0043] Fig.13 Schematic diagram of parameter estimation accuracy of the frequency-frequency diagram of the frequency-hopping signal of one of the dual-network stations after denoising. DETAILED DESCRIPTION

[0044] The present invention will be further described in detail below in conjunction with embodiments and drawings, but the embodiments of the present invention are not limited thereto.

[0045] Example 1

[0046] like Figures 1 to 13 As shown, this mainstream frequency hopping signal parameter estimation method in the background technology cannot estimate the instantaneous bandwidth of the frequency hopping signal, so the inventor of the present invention estimates the parameters such as the hopping time, hopping speed, and frequency hopping period of the frequency hopping signal from another perspective of image processing. In particular, this method of estimating the parameters of the frequency hopping signal using image processing can estimate the instantaneous bandwidth parameters of the frequency hopping signal, which is impossible for the previous mainstream method to do. And because this method is particularly suitable for the situation where the time-frequency diagram is separated, this method is more applicable than the original method. It is not only limited to identifying frequency hopping signals, but can also be applied to various types of signals separated by spectrum. A corresponding adaptive thresholded binary clustering algorithm is proposed for the problem of removing noise in the frequency hopping signal parameter estimation based on image processing, which greatly improves the parameter estimation accuracy of the frequency hopping signal in a noisy environment.

[0047] Aiming at the problem of low parameter estimation accuracy of frequency hopping signals under low signal-to-noise ratio, an adaptive fixed-threshold binary clustering algorithm suitable for image processing denoising is proposed, and experimental simulation results show that this method can greatly improve the parameter estimation accuracy.

[0048] The present invention discloses a frequency hopping signal processing method based on binary classification and clustering, and belongs to the field of signal processing. Aiming at the problem of low accuracy of frequency hopping signal parameter estimation under low signal-to-noise ratio in engineering projects, a frequency hopping signal processing method based on binary classification and clustering is proposed to improve the accuracy of frequency hopping parameter estimation. The method comprises the following steps: 1) generating and collecting frequency hopping signal data; 2) denoising the frequency hopping signal using an adaptive fixed threshold binary clustering algorithm; 3) using the denoised frequency hopping signal data for frequency hopping signal parameter estimation; 4) testing the performance improvement after using the method.

[0049] A frequency hopping signal processing method based on binary classification clustering, several key steps in the adaptive threshold binary classification clustering algorithm, such as Figure 1 As shown:

[0050] Step S1: Represent the frequency hopping signal in the form of discrete points in the complex domain;

[0051] Step S2: Separate the signal class and the noise class by finding the centroid;

[0052] Step S3: calculating the modulus of discrete points in the complex domain of the signal type and the noise type and representing them on the coordinate axis;

[0053] Step S4: extract the signal class by multiplying the corresponding grayscale values ​​of the points of the signal class and the noise class on the coordinate axis by 1 and 0 respectively, thereby realizing denoising processing on the frequency hopping signal.

[0054] 1. Adaptive threshold binary clustering algorithm:

[0055] like Figures 2 to 8 As shown, Figures 2 to 8 The iterative process of the adaptive fixed threshold bisection clustering algorithm in the complex domain is demonstrated; wherein the x-axis represents the real axis, the y-axis represents the imaginary axis, the x-axis unit is: +1, and the y-axis unit is: +j.

[0056] Figures 2 to 7 They are the step-by-step diagrams of the iterative process. Figure 2 The green dots in the middle represent sampling points. Figure 3 In the above figure, we artificially select an initial signal cluster center, which is indicated by a blue cross, and artificially select an initial noise cluster center, which is indicated by a red cross. If the distance from one of the sampling points to the signal cluster center is closer than the distance from the sampling point to the noise cluster center, then the sampling point is classified as a signal and the color of the sampling point is changed to blue. Otherwise, the sampling point is classified as a noise class and the color of the sampling point is changed to red, so we get Figure 4 . Figure 4 In the figure, the centroid of the sampling points marked in blue is found, and then the centroid is updated to the new signal cluster center and represented by a blue cross. The centroid of the sampling points marked in red is found, and then the centroid is updated to the new noise cluster center and represented by a red cross, so we get Figure 5 . Figure 5 If the distance between one of the sampling points and the signal cluster center is closer than the distance between the sampling point and the noise cluster center, then the sampling point is classified as a signal and the sampling point is changed to blue. Otherwise, the sampling point is classified as a noise class and the sampling point is changed to red, so we get Figure 6 . Figure 6 Repeat the previous steps to get Figure 7 If the above process only displays one result, then Figure 8 shown.

[0057] like Fig. 9 As shown, Fig. 9 This is the effect diagram of the adaptive fixed threshold bisection clustering algorithm in the complex domain (the x-axis represents the real axis, the y-axis represents the imaginary axis, the x-axis unit is: +1, and the y-axis unit is: +j).

[0058] Assumptions Fig. 9 The original data in the complex domain is obtained after sampling the frequency hopping signal, which contains two types of sampling point data: frequency hopping signal and noise signal. After multiple iterations, the frequency hopping signal data and the noise data are separated by clustering, and the clustering centers of the frequency hopping signal and the noise signal are obtained respectively.

[0059] pass Fig. 9 By finding the modulus value of each sampling point data in the complex domain, we can get the grayscale value in the image of each sampling point data, and mark all the grayscale values ​​on the coordinate axis, then we can get the following figure, as shown in Fig.10 shown. Fig.10 This is the effect diagram of the adaptive threshold binary clustering algorithm (the x-axis represents the grayscale value in the image, the y-axis has no meaning, the x-axis unit is +1, and the y-axis has no unit).

[0060] First of all, it should be clear that in the time-frequency diagram, we take the grayscale values ​​of all points as the objects of classification, and we can take the values ​​of signal and noise as two categories, called signal class and noise class respectively. In the adaptive threshold binary clustering algorithm, the two initial points can be randomly selected, but through multiple random selections of the values ​​of the two initial points, we found that the initial point of the noise class is the lowest point of the grayscale value and the initial point of the signal class is the highest point of the grayscale value. In this way, after the adaptive threshold binary clustering, we can separate the signal and noise more quickly and accurately, that is, find the cluster center of the signal as L1 and the cluster center of the noise as L2; ​​then the correction threshold can be set to th2 = (L1 + L2) / 2. In the time-frequency diagram, the grayscale values ​​less than the threshold are set to 0; the grayscale values ​​greater than or equal to the threshold are retained. The most critical thing in this process is to find different threshold values ​​th2 corresponding to different signal-to-noise ratios through the adaptive threshold binary clustering algorithm, so as to complete the denoising process under different signal-to-noise ratios. This adaptive determination of different threshold values ​​corresponding to different signal-to-noise ratios can actually adaptively denoise noises of different strengths, greatly improving the denoising effect. This adaptive, or automatic, determination of the threshold is the core innovation of this algorithm.

[0061] like Fig.11 As shown, Fig.11 The comparison of the time-frequency diagram of the frequency hopping signal of multiple network stations before and after denoising (left figure. Before denoising; right figure. After denoising; the x-axis represents time, in seconds; the y-axis represents frequency, in Hz).

[0062] When the signal-to-noise ratio is -2dB, the frequency hopping signals of the two stations are networked, where the dwell time of each hop of the frequency hopping signal of the first station is set to twice the dwell time of each hop of the frequency hopping signal of the second station, and then the time-frequency waterfall diagram of the frequency hopping signals of the two stations is generated, as shown in FIG. Fig.11As shown in the left figure. Then, the frequency hopping signals of the two stations are denoised by the adaptive threshold binary clustering algorithm and then the time-frequency waterfall diagram is generated. In this way, the frequency hopping signals of the two stations are separated from the noise in the environment according to the size of the image grayscale value, and the grayscale value of the separated frequency hopping signals of the two stations is multiplied by 1, and the grayscale value of the separated environmental noise signal is multiplied by 0, as shown in Fig.11 As shown in the right figure. The time-frequency waterfall diagram of the frequency hopping signals of the two network stations obtained after denoising by the adaptive threshold binary clustering algorithm is clearer and more accurate than the time-frequency waterfall diagram of the frequency hopping signals of the two network stations without denoising by the adaptive threshold binary clustering algorithm. This makes preliminary preparations for the subsequent image processing process.

[0063] 2. Use the denoised frequency hopping signal data for frequency hopping signal parameter estimation:

[0064] The SSD detection framework for image processing used in the present invention is implemented through Python source code and can be obtained on Cornell University's official website through its open source license (https: / / github.com / weiliu89 / caffe / tree / ssd). In 2017, scholars such as Shaoqing Ren and Ross Girshick proposed the Fast RCNN detection framework, which significantly improved the recognition accuracy and recognition speed of the RCNN detection framework. In order to solve a series of problems in RCNN generating candidate regions, scholars such as Joseph Redmon proposed the YOLO target monitoring framework. In 2016, scholars such as Wei Liu and DragomirAnguelov proposed the SSD detection framework, which not only adopted the anchoring mechanism of FastRCNN, but also inherited the regression strategy of YOLO. The main idea of ​​the SSD detection framework is to use convolution kernels to predict the type and offset coordinates of the detection target on feature maps of different sizes, which greatly improves the accuracy and real-time performance of target detection. The SSD detection framework model is based on the multi-box target theory, which can be used to train multiple targets.

[0065] The SSD detection framework can return the normalized coordinates (y1, x1, y2, x2) of each frequency hopping segment of the specified network station detected in the time-frequency waterfall diagram. Then, the center frequency and instantaneous bandwidth of the frequency hopping signal can be obtained by multiplying the detected normalized coordinates by the sampling frequency in proportion. Similarly, the residence time of each hop can be obtained by multiplying the detected normalized coordinates by the sampling time in proportion, as shown in the following formula:

[0066] f=[(y2+y1) / 2]f s ,

[0067] B=(y2-y1)*f s,

[0068] R=1 / [(x2-x1)*T s ];

[0069] Fig.12 and Fig.13 The result of frequency hopping signal parameter estimation after denoising by adaptive fixed threshold binary classification algorithm.

[0070] Fig.12 Schematic diagram of parameter estimation of the frequency-frequency diagram of the frequency-hopping signal of one of the dual-network stations after denoising (the x-axis represents time in seconds; the y-axis represents frequency in Hz).

[0071] Fig.13 The accuracy of parameter estimation of the frequency-frequency diagram of the frequency-hopping signal of one of the two networks after denoising is demonstrated.

[0072] The present invention proposes a corresponding adaptive threshold binary clustering algorithm for the problem of removing noise in frequency hopping signal parameter estimation based on image processing, which greatly improves the parameter estimation accuracy of the frequency hopping signal in a noisy environment.

[0073] As described above, the present invention can be preferably implemented.

[0074] All features disclosed in all embodiments in this specification, or steps in all methods or processes implicitly disclosed, except for mutually exclusive features and / or steps, can be combined and / or expanded or replaced in any manner.

[0075] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. According to the technical essence of the present invention, within the spirit and principles of the present invention, any simple modification, equivalent replacement and improvement made to the above embodiment still falls within the protection scope of the technical solution of the present invention.

Claims

1. A frequency hopping signal processing method based on binary classification, characterized in that: Adopting adaptive threshold binary clustering algorithm to separate the signal data and noise data of frequency hopping signal, thus denoising the frequency hopping signal; The following steps are involved: S1, complex domain representation: the frequency hopping signal is represented as discrete points in the complex domain; S2, separation: separating the signal data and noise data of the frequency hopping signal; S3, modulus representation: the discrete points in the complex domain of the signal data and the noise data are expressed on the coordinate axis after the modulus is calculated; S4, signal data extraction: extracting the signal data, thereby realizing denoising of the frequency hopping signal; In step S3, the modulus value of each point of the sampling point data in the complex domain is calculated to obtain the grayscale value in the image of each sampling point data, and then all the grayscale values ​​are marked on the coordinate axis; In step S4, the signal data is extracted by multiplying the corresponding grayscale values ​​of the points of the signal data on the coordinate axis by 1 or 0. The method is as follows: find the cluster center of the signal as L1 and the cluster center of the noise as L2, set the correction threshold to th2 = (L1 + L2) / 2, set the grayscale values ​​< the correction threshold to 0 in the time-frequency diagram, retain the grayscale values ​​≥ the correction threshold, multiply the grayscale values ​​of the separated frequency-hopping signals of the two network stations by 1, and multiply the grayscale value of the separated environmental noise signal by 0.

2. A method for processing frequency hopping signals based on binary classification according to claim 1, characterized in that: In step S2, the signal data and noise data of the frequency hopping signal are separated by a centroid iteration method, and the clustering centers of the frequency hopping signal and the noise signal are obtained respectively.

3. The method for processing frequency hopping signals based on binary classification according to claim 2, characterized in that: The following steps are also included: S5, parameter estimation: using the denoised frequency hopping signal data for frequency hopping signal parameter estimation.

4. The method for processing frequency hopping signals based on binary classification according to claim 3, characterized in that: In step S5, the normalized coordinates (y1, x1, y2, x2) of each frequency hopping segment are first obtained, and then the center frequency and instantaneous bandwidth of the frequency hopping signal are obtained in proportion according to the normalized coordinates, and the dwell time of each hop is obtained, as shown in the following formula: f=[(y2+y1) / 2]f s , B=(y2-y1)*f s , R=1 / [(x2-x1)*T s ]; Among them, f is the center frequency of the frequency hopping signal obtained by proportion according to the normalized coordinates, y1 is the lower boundary of each frequency hopping band, y2 is the upper boundary of each frequency hopping band, and f s is the sampling frequency of the frequency hopping signal, B is the instantaneous bandwidth of the frequency hopping signal obtained by proportionality based on the normalized coordinates, x1 is the left boundary of each frequency hopping segment, x2 is the right boundary of each frequency hopping segment, R is the dwell time of each hop, T s =1 / f s .

5. A frequency hopping signal denoising system based on adaptive threshold binary classification, characterized in that: A method for processing a frequency hopping signal based on binary classification clustering according to any one of claims 1 to 4, comprising the following modules connected in sequence: Complex domain representation module: used to represent the frequency hopping signal in the complex domain as discrete points; Separation module: used to separate the signal data and noise data of the frequency hopping signal; Modulus value representation module: used to calculate the modulus value of discrete points in the complex domain of signal data and noise data and represent them on the coordinate axis; Signal data extraction module: used to extract signal data, so as to realize denoising processing of frequency hopping signal.

6. The frequency hopping signal denoising system based on adaptive threshold binary classification according to claim 5 is characterized in that: It also includes a parameter estimation module connected to the signal data extraction module, and the parameter estimation module is used for using the denoised frequency hopping signal data for frequency hopping signal parameter estimation.

Images