Signal blind detection method based on hidden Markov model variational mode decomposition

Abstract

The invention provides a signal blind detection method based on hidden Markov model variational mode decomposition. The signal blind detection method comprises the following steps: obtaining a boundary detection result of a received signal based on a hidden Markov model algorithm; determining the number of sub-band signals in the received signal and the range of each sub-band signal based on a variational mode decomposition model algorithm; hidden Markov model algorithm detection results in the existence range of each sub-band signal are combined, and then signal components and non-signal components in the detection results are effectively distinguished by using a plurality of intrinsic mode function IMF marginal spectrums; according to the method, the USB measurement and control signal can be correctly detected in the received signal, the method can be used in a non-cooperative communication scene and is not influenced by wavelet basis selection, the number of wavelet layers and a noise interference filtering threshold value, most of signal components of the USB measurement and control signal can be detected, and a complete USB measurement and control signal can be prevented from being detected into a plurality of sub-band signals.

Description

Technical Field

[0001] This invention belongs to the field of satellite signal processing and relates to a blind detection method for receiving downlink unified S-band (Unified S Band, USB) telemetry and control signals transmitted by satellites on the ground in practical applications. It includes detecting the number of existing sub-band signals and the frequency position of each signal. Specifically, it refers to a signal blind detection method based on variational mode decomposition of hidden Markov models. Background Technology

[0002] With the continuous development of aerospace technology, the number of on-orbit targets and maneuvering vehicles in space has increased dramatically, especially with large low-Earth orbit satellite constellations such as Starlink and OneWeb, leading to a greater complexity of the space environment. Given the large number of space targets, the downlink signals received from satellites on the ground typically contain multiple sub-band signals. For signal analysis tasks, determining the number of sub-band signals in the received signal and locating the position of each sub-band signal in the received signal spectrum is paramount. Only after locating the signal can subsequent work such as parameter analysis proceed. In cooperative communication scenarios, since the receiver knows the satellite signal transmission pattern, frequency, and bandwidth, a simple bandpass filter can usually obtain the desired signal. However, in non-cooperative communication scenarios, since the frequency and bandwidth of the signal are unknown, it is necessary to detect and separate each sub-band signal from the numerous received signals. Furthermore, fluctuating noise floor and some special modulation methods can also affect the estimation of the number of sub-band signals in the received signal. Therefore, correctly detecting each sub-band signal in non-cooperative scenarios is a challenging problem.

[0003] Existing blind signal detection algorithms mainly fall into three categories: energy detection, edge detection, and hidden state detection. The performance of energy detection algorithms is affected by the selection of threshold values. For example, the classic localization algorithm based on double thresholding-adjacent cluster combining (LAD-ACC) requires prior knowledge of the signal-to-noise ratio (SNR). Edge detection algorithms mainly include wavelet edge detection and Hilbert edge detection. Both methods detect transitions at signal boundaries. The performance of wavelet edge detection is affected by the selected wavelet basis and the number of decomposition levels, while the performance of Hilbert edge detection is affected by the threshold value set to filter out noise interference. Hidden state detection primarily uses Hidden Markov Models (HMMs) to estimate hidden states, thus determining whether a signal is present or absent. The advantage of this method is that it does not require any prior information about the signal, making it suitable for non-cooperative communication scenarios.

[0004] The methods described above achieve good detection results for general communication signals. The energy detection algorithm is mainly affected by the threshold selection, while the edge detection and hidden state detection algorithms are affected by the power spectrum frequency resolution. However, due to the unique modulation method of USB monitoring and control signals, their spectrum exhibits spectral separation characteristics, leading to a sharp decline in the performance of existing detection algorithms. The LAD-ACC and HMM algorithms both suffer from significant signal count estimation errors (detecting a complete USB monitoring and control signal as multiple independent sub-band signals), but they can generally cover most signal components. Edge detection algorithms, on the other hand, have significant bandwidth positioning errors (detecting only the most central component of the USB monitoring and control signal), but do not have significant errors in signal count. These numerous shortcomings of existing detection algorithms present new challenges to the detection of USB monitoring and control signals. Summary of the Invention

[0005] To address the problem of detecting USB monitoring and control signals in non-cooperative scenarios, and considering that Hidden Markov Model (HMM) algorithms do not require any prior information, this invention proposes a blind signal detection method based on Hidden Markov Model Variational Mode Decomposition (VMD). The VMD algorithm is introduced to achieve blind detection of the received signal.

[0006] Therefore, the present invention adopts the following technical solution: a blind signal detection method based on hidden Markov model variational mode decomposition, comprising the following steps: Step 1: Obtain the boundary detection results of the received signal based on the Hidden Markov Model algorithm; Step 2: Determine the number of sub-band signals and the range of each sub-band signal in the received signal based on the variational mode decomposition model algorithm; In step 2, determining the number of sub-band signals and the range of each sub-band signal in the received signal involves using a variational mode decomposition algorithm to decompose the signal into multiple intrinsic mode functions (IMFs). At this point, the received signal... It can be represented as: in, K The number of decomposition layers, To sum all modes so that they can be reconstructed into the decomposed signal ; Step 3: Merge the detection results of the Hidden Markov Model algorithm within the range of each sub-band signal, and then use multiple intrinsic mode functions (IMF) marginal spectra to effectively distinguish the signal components and non-signal components in the detection results; Step 2 further includes adaptively determining the optimal decomposition level for the number of sub-band signals based on a stationary blind clustering model algorithm, comprising the following steps: First, a range of values ​​for the decomposition level is preset. By traversing different decomposition levels from small to large, the marginal spectrum corresponding to the variational mode decomposition model algorithm at each decomposition level is obtained. Then, blind clustering is performed on the maxima in the marginal spectrum at each decomposition level. Based on the energy distribution of each sub-band signal within the received signal, a binary classification method or a multi-class classification method is used to obtain the optimal decomposition level. Finally, the variational mode decomposition model algorithm is used again to obtain the marginal spectrum result at that level to determine the actual number of sub-band signals.

[0007] Furthermore, the variational mode decomposition model algorithm is expressed as follows: ,in Represents the set of all modalities. This represents the set of center frequencies corresponding to all modes, and the symbol "*" indicates the convolution operation. k This represents the current decomposition level. To sum all modes so that they can be reconstructed into the decomposed signal The signal is ultimately decomposed. It can be represented as: The settings within K The total number of decomposition levels is obtained when... K After each IMF, there exists a residual term. .

[0008] Furthermore, the binary classification method is the DBSACAN method, which considers the larger class of maxima as signal components, counts the number of maxima in that class, and records it as the number of signals.

[0009] Furthermore, the multi-classification method is the DBSACAN method, which considers the class with the most extreme points as a non-signal component, and each of the remaining classes represents a signal component.

[0010] Furthermore, in step 2, the method for determining the optimal number of decomposition layers includes: finding the intervals of decomposition layers corresponding to the number of all non-zero and consecutively equal signals, selecting the longest interval, finding the minimum point among the variances of multiple intrinsic mode function (IMF) marginal spectrum minimum points, and taking the number of layers corresponding to this point as the optimal number of layers.

[0011] The beneficial effects of this invention are as follows: The USB measurement and control signal blind detection algorithm proposed in this invention first detects the signal components of each sub-band of the received signal using a Hidden Markov Model (HMM) algorithm. Then, it decomposes the received signal using a Variational Mode Decomposition (VM) algorithm to obtain multiple Integral Multi-Function (IMF). The number of actual sub-band signals is determined by the marginal spectrum of the IMF. Since the decomposition result of the VM algorithm is affected by the number of decomposition layers, this invention also proposes a stationary blind clustering algorithm to adaptively determine the optimal number of decomposition layers for the VM algorithm. Under the optimal number of decomposition layers, it can be ensured that the number of sub-band signals determined by the VM algorithm is closer to the true result.

[0012] Compared to traditional signal detection algorithms, this invention can correctly detect USB monitoring and control signals in received signals. Compared to the LADACC algorithm, the proposed algorithm does not require any prior information about the signal, can be used in non-cooperative communication scenarios, and can detect a complete USB monitoring and control signal. Compared to wavelet edge detection algorithms and Hilbert edge detection algorithms, it is not affected by wavelet basis selection, wavelet layer number, and noise interference filtering threshold, has fewer limiting factors for detection performance, and can detect most of the signal components of the USB monitoring and control signal. Compared to Hidden Markov Model (HMM) algorithms, it can avoid detecting a complete USB monitoring and control signal as multiple sub-band signals. At the same time, it can, to some extent, avoid the false boundaries caused by noise floor fluctuations or interference detected by HMM algorithms. Attached Figure Description

[0013] Figure 1 This is a flowchart of the signal blind detection method based on Hidden Markov Model Variational Mode Decomposition of the present invention; Figure 2 This is an image of the signal number-normalized power spectral density determined using the variational mode decomposition model algorithm in an embodiment of the present invention; Figure 3 This is an image of the marginal spectrum of the signal number determined using the variational mode decomposition model algorithm in an embodiment of the present invention. Detailed Implementation

[0014] The present invention will be further described and illustrated below with reference to the accompanying drawings and specific embodiments.

[0015] like Figure 1 As shown, a blind signal detection method based on Hidden Markov Model Variational Mode Decomposition includes the following steps: Step 1: Obtain the boundary detection results of the received signal based on the Hidden Markov Model algorithm; Step 2: Determine the number of sub-band signals and the range of each sub-band signal in the received signal based on the variational mode decomposition model algorithm; In step 2, determining the number of sub-band signals and the range of each sub-band signal in the received signal involves using a variational mode decomposition algorithm to decompose the signal into multiple intrinsic mode functions (IMFs). At this point, the received signal... It can be represented as: in, K The number of decomposition layers, To sum all modes so that they can be reconstructed into the decomposed signal ; Step 3: Merge the detection results of the Hidden Markov Model algorithm within the range of each sub-band signal, and then use multiple intrinsic mode functions (IMF) marginal spectra to effectively distinguish the signal components and non-signal components in the detection results; Step 2 further includes adaptively determining the optimal decomposition level for the number of sub-band signals based on a stationary blind clustering model algorithm, comprising the following steps: First, a range of values ​​for the decomposition level is preset. By traversing different decomposition levels from small to large, the marginal spectrum corresponding to the variational mode decomposition model algorithm at each decomposition level is obtained. Then, blind clustering is performed on the maxima in the marginal spectrum at each decomposition level. Based on the energy distribution of each sub-band signal within the received signal, a binary classification method or a multi-class classification method is used to obtain the optimal decomposition level. Finally, the variational mode decomposition model algorithm is used again to obtain the marginal spectrum result at that level to determine the actual number of sub-band signals.

[0016] In the above technical solution, the variational mode decomposition model algorithm is expressed as follows: ,in Represents the set of all modalities. This represents the set of center frequencies corresponding to all modes, and the symbol "*" indicates the convolution operation. k This represents the current decomposition level. To sum all modes so that they can be reconstructed into the decomposed signal The signal is ultimately decomposed. It can be represented as: The settings within K The total number of decomposition levels is obtained when...K After each IMF, there exists a residual term. .

[0017] In the above technical solution, the binary classification method is the DBSACAN method, which considers the larger class of maxima as signal components, counts the number of maxima in that class, and records it as the number of signals.

[0018] In the above technical solution, the multi-classification method is the DBSACAN method, which considers the class with the most extreme points as a non-signal component, and each of the remaining classes represents a signal component.

[0019] In the above technical solution, step 2, the method for determining the optimal number of decomposition layers includes: finding the interval of decomposition layers corresponding to the number of all non-zero and consecutively equal signals, selecting the longest interval, finding the minimum point among the variances of multiple intrinsic mode function (IMF) marginal spectrum minimum points, and taking the number of layers corresponding to this point as the optimal number of layers.

[0020] It should be noted that determining the number of sub-band signals and the approximate range of each sub-band signal in the received signal using the variational mode decomposition model algorithm essentially involves first decomposing the signal using the variational mode decomposition model algorithm to obtain multiple intrinsic mode functions (IMFs). At this point, the received signal... It can be approximated as: in K To determine the number of decomposition layers. For a single IMF, it can be represented by the Hilbert transform as an expression of instantaneous amplitude and instantaneous frequency: The above formula is also called The Hilbert spectrum can then be obtained. The Hilbert spectrum is as follows: Finally, the Hilbert spectrum is accumulated in the time domain to obtain the marginal spectrum, i.e.: In the marginal spectrum, the number of signals and their range are determined by identifying convex components. Specifically, the number of signals is estimated using the number of convex components, and the range is determined based on the locations of the minimum points of these convex components. Taking a received signal containing four sub-band signals as an example, this demonstrates how to determine the number of signals using the marginal spectrum (VMD decomposition layer number 10, penalty factor 1000): Based on... Figure 3 It can be seen that the variational mode decomposition model algorithm yields a total of 10 IMFs, of which four components have significantly higher energies than the others. These four components are exactly the same as... Figure 2 The four sub-band signals correspond to each other, and the frequency range between the minimum points on both sides of these four IMF components is the approximate frequency range in which the four sub-band signals exist.

[0021] The process of determining the number of sub-band signals using the variational mode decomposition (VM) model algorithm is functionally equivalent to using a false alarm threshold to determine the detection results of a hidden Markov model (HMM) algorithm. The difference lies in the fact that traditional false alarm threshold decision-making first calculates a threshold based on a preset false alarm probability, and then classifies components exceeding the threshold as signal components. In contrast, the VM model algorithm effectively distinguishes between signal and non-signal components by utilizing the characteristic that only signal components are highlighted in the IMF marginal spectrum. Based on the bandwidth range of the signal component, most false alarm components can be filtered out. Furthermore, this method can effectively avoid misclassifying low-energy signal components as false alarm components due to improper threshold settings.

[0022] Furthermore, it's important to note that the decomposition results of the variational mode decomposition (VMD) algorithm are affected by the number of decomposition layers and the penalty factor. The number of decomposition layers affects the number of IMFs in the marginal spectrum and the bandwidth of each IMF, while the penalty factor only affects the bandwidth of each IMF. In reality, the penalty factor's value has a much smaller impact on bandwidth than the number of decomposition layers. At the same number of decomposition layers, a change in the penalty factor of at least 1000 is needed to significantly alter the bandwidth of the IMFs in the marginal spectrum, while even a single increase or decrease in the number of decomposition layers is sufficient to produce a significant change in the marginal spectrum.

[0023] Given that the number of decomposition layers is a major influencing factor, and that an appropriate number of decomposition layers helps the variational mode decomposition model algorithm better determine the number of sub-band signals present in the received signal, this invention also proposes a stationary blind clustering algorithm to adaptively determine the optimal number of decomposition layers in the variational mode decomposition model algorithm. The principle of this algorithm is as follows: (1) Preset a range of values ​​for the number of decomposition layers, and obtain the marginal spectrum of the variational mode decomposition model algorithm under each decomposition layer by traversing different decomposition layers from small to large. (2) Blind clustering is performed on the maxima in the marginal spectrum at each decomposition level, and two different strategies are adopted according to the energy distribution of each sub-band signal in the received signal: (3) If the signal energy distribution of each sub-band is relatively uniform, a binary classification method (such as the K-medoids method) is adopted. The larger maximum value is considered as a signal component, and the number of maximum values ​​in this class is counted and recorded as the number of signals. (4) If the energy distribution of each subband signal is significantly different, a multi-classification method (such as the DBSACAN method) is adopted, and the class with the most extreme points is considered as a non-signal component, while each of the remaining classes represents a signal component.

[0024] (5) Find the interval of decomposition level corresponding to the number of all non-zero and consecutive equal signals, and select the longest interval (if there are intervals of the same length, select the interval with the larger decomposition level). It is believed that the decomposition result in this interval is stable and closer to the true result. (6) Within the interval obtained in the previous step (including the boundary of the interval), find the minimum point with the smallest variance among the minimum points of the marginal spectrum (if this condition is not met, select the minimum point). At this time, it is considered that the IMF distribution obtained by decomposition is relatively uniform, and the number of layers corresponding to this point is taken as the optimal number of layers.

[0025] This algorithm operates on specific signals under current conditions, rather than a fixed signal model. Therefore, it can adapt to different signal types, signal-to-noise ratio environments, and varying frequency resolutions.

[0026] After obtaining the optimal number of decomposition layers, the variational mode decomposition model algorithm is used again to obtain the marginal spectrum results under that number of layers to determine the actual number of sub-band signals.

[0027] The USB measurement and control signal blind detection algorithm proposed in this invention first detects the sub-band signal components of the received signal using a Hidden Markov Model (HMM) algorithm. Then, it decomposes the received signal using a Variational Mode Decomposition (VM) algorithm to obtain multiple Integrated Motion Components (IMFs). The number of actual sub-band signals is determined by the marginal spectrum of the IMFs. Since the decomposition result of the VM algorithm is affected by the number of decomposition layers, this invention also proposes a stationary blind clustering algorithm to adaptively determine the optimal number of decomposition layers for the VM algorithm. At the optimal number of decomposition layers, the number of sub-band signals determined by the VM algorithm is closer to the true result.

[0028] Compared to traditional signal detection algorithms, this invention can correctly detect USB monitoring and control signals in received signals. Compared to the LADACC algorithm, the proposed algorithm does not require any prior information about the signal, can be used in non-cooperative communication scenarios, and can detect a complete USB monitoring and control signal. Compared to wavelet edge detection algorithms and Hilbert edge detection algorithms, it is not affected by wavelet basis selection, wavelet layer number, and noise interference filtering threshold, has fewer limiting factors for detection performance, and can detect most of the signal components of the USB monitoring and control signal. Compared to Hidden Markov Model algorithms, it can avoid detecting a complete USB monitoring and control signal as multiple sub-band signals. At the same time, it can, to a certain extent, avoid the detection of false boundaries caused by noise floor fluctuations or interference by Hidden Markov Model algorithms.

Claims

1. A blind signal detection method based on Hidden Markov Model Variational Mode Decomposition, characterized in that, Includes the following steps: Step 1: Obtain the boundary detection results of the received signal based on the Hidden Markov Model algorithm; Step 2: Determine the number of sub-band signals and the range of each sub-band signal in the received signal based on the variational mode decomposition model algorithm; In step 2, determining the number of sub-band signals and the range of each sub-band signal in the received signal involves using a variational mode decomposition algorithm to decompose the signal into multiple intrinsic mode functions (IMFs). At this point, the received signal... It can be represented as: in, K The number of decomposition layers, To sum all modes so that they can be reconstructed into the decomposed signal ; Step 3: Merge the detection results of the Hidden Markov Model algorithm within the range of each sub-band signal, and then use multiple intrinsic mode functions (IMF) marginal spectra to effectively distinguish the signal components and non-signal components in the detection results; Step 2 further includes adaptively determining the optimal decomposition level for the number of sub-band signals based on a stationary blind clustering model algorithm, comprising the following steps: First, a range of values ​​for the decomposition level is preset. By traversing different decomposition levels from small to large, the marginal spectrum corresponding to the variational mode decomposition model algorithm at each decomposition level is obtained. Then, blind clustering is performed on the maxima in the marginal spectrum at each decomposition level. Based on the energy distribution of each sub-band signal within the received signal, a binary classification method or a multi-class classification method is used to obtain the optimal decomposition level. Finally, the variational mode decomposition model algorithm is used again to obtain the marginal spectrum result at that level to determine the actual number of sub-band signals.

2. The signal blind detection method based on Hidden Markov Model Variational Mode Decomposition according to claim 1, characterized in that, The variational mode decomposition model algorithm is expressed as follows: ,in Represents the set of all modalities. This represents the set of center frequencies corresponding to all modes, and the symbol "*" represents the convolution operation. k This represents the current decomposition level. To sum all modes so that they can be reconstructed into the decomposed signal The signal is ultimately decomposed. It can be represented as: The settings within K The total number of decomposition levels is obtained when... K After each IMF, there exists a residual term. .

3. The signal blind detection method based on Hidden Markov Model Variational Mode Decomposition according to claim 1, characterized in that, The binary classification method is the DBSACAN method, which considers the larger class of maxima as signal components, counts the number of maxima in that class, and records it as the number of signals.

4. The signal blind detection method based on Hidden Markov Model Variational Mode Decomposition according to claim 1, characterized in that, The multi-classification method is the DBSACAN method, which considers the class with the most extreme points as a non-signal component, and each of the remaining classes represents a signal component.

5. The signal blind detection method based on Hidden Markov Model Variational Mode Decomposition according to claim 1, characterized in that, In step 2, the method for determining the optimal number of decomposition layers includes: finding the interval of decomposition layers corresponding to the number of all non-zero and consecutively equal signals, selecting the longest interval, finding the minimum point among the variances of the minimum points of the marginal spectrum of multiple intrinsic mode functions (IMFs), and taking the number of layers corresponding to this point as the optimal number of layers.

Images