An improved synchronous extraction time-frequency analysis method based on adaptive window length
Through the improved synchronous extraction time-frequency analysis method of adaptive window length, the problem of insufficient performance of fuzzy representation and multi-component signal analysis in time-varying signal processing by traditional time-frequency analysis methods is solved, and efficient time-frequency resolution and signal are achieved Retention of strength information is suitable for a variety of signal forms.
Patent Information
- Application Number
- CN202110657730.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-06-14
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2041-06-14
AI Technical Summary
Traditional time-frequency analysis methods have fuzzy time-frequency representations when processing time-varying signals, making it difficult to provide an accurate time-frequency description, especially for multi-component signals and emphasize frequency signals, and the analysis performance of adaptive parameters is insufficient, and adaptive parameters are difficult to determine.
Using an improved synchronous extraction time-frequency analysis method based on adaptive window length, the frequency change rate is estimated through short-time Fourier transform and second-order local modulation operator, the peeling algorithm searches for local energy peaks, designs adaptive time-frequency windows, and constructs a new short-time Fourier transform, improves time-frequency resolution and preserves signal strength information.
It realizes clear time-frequency representation, adapts to various signal forms, improves time-frequency resolution and energy aggregation, and can better extract weak amplitude signals and retain the intensity information of each component.
Smart Images

Figure CN113553901B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to signal processing technology, and in particular relates to an improved synchronous extraction time-frequency analysis method based on adaptive window length. Background Art
[0002] Time-frequency analysis (TFA) is an effective tool for describing the time-varying characteristics of non-stationary signals and has been widely used in engineering fields such as communications, radar, and sonar. Classical linear methods, such as short-time Fourier transform (STFT) and wavelet transform (WT), can expand one-dimensional time series signals to a two-dimensional time-frequency (TF) plane, from which the time-varying characteristics of the signal can be observed and the decomposition of multi-component signals can be achieved. However, limited by the Heisenberg uncertainty principle, the time-frequency representation generated by traditional methods is usually fuzzy and cannot provide an accurate time-frequency description for time-varying signals.
[0003] The shortcomings of traditional TFA methods have severely limited their application in practical data processing. To solve this problem, many advanced post-processing methods have been developed in recent years, such as the reallocation method (RM), synchronous compression transform (SST), and synchronous extraction transform (SET). RM first calculates the reallocated position of each time-frequency point based on the time-frequency phase information, and then redistributes the time-frequency coefficients to the instantaneous trajectory along the two-dimensional time-frequency direction, so it can greatly improve the time-frequency resolution, but RM lacks the ability to reconstruct the signal. SST only squeezes the time-frequency coefficients into the instantaneous trajectory in the frequency direction, so it can accurately reconstruct the signal while improving the performance of the classical TFA method. Compared with SST, SET only retains the time-frequency coefficients that are most relevant to the time-varying characteristics of the signal, and has higher energy concentration and noise robustness. The above post-processing methods are all based on STFT with a fixed window length, and the analysis performance of strong frequency signals and multi-component signals is difficult to be satisfactory.
[0004] The Chinese patent with publication number CN 107576943A discloses an adaptive time-frequency synchronization compression method based on Rayleigh entropy. The method first uses Rayleigh entropy to measure the aggregation of local energy of the signal, estimates the optimal window parameter sequence suitable for signal time-frequency analysis, and improves the aggregation through time-frequency synchronization compression to achieve the separation of multi-component signals. The method has a high time-frequency resolution, and the adaptive time window can also adapt to more forms of signals, but for multi-component signals with large frequency changes, it is difficult to adapt to the changes of each component at the same time.
[0005] A Chinese patent with publication number CN108694392A discloses a high-precision synchronous extraction generalized S transform time-frequency analysis method. This method combines the three-parameter generalized S transform with synchronous extraction, can adjust the window length according to the signal frequency, and has a high time-frequency resolution. However, its performance is greatly affected by the parameters, and it is difficult to determine the optimal parameters. It is not suitable for strong frequency signals and loses signal strength information. Summary of the invention
[0006] The object of the present invention is to provide an improved synchronization extraction time-frequency analysis method based on adaptive window length.
[0007] The technical solution to achieve the purpose of the present invention is: an improved synchronous extraction time-frequency analysis method based on adaptive window length, comprising the following steps:
[0008] Step 1: Perform short-time Fourier transform on the input signal;
[0009] Step 2, estimating the frequency change rate of each time-frequency point according to the second-order local modulation operator;
[0010] Step 3, iteratively solve the optimization problem through the stripping algorithm, search for local energy peaks, extract time-frequency ridges, estimate the number of signal components, and obtain the instantaneous frequency change rate of each component;
[0011] Step 4: For a single-component signal, an adaptive time window is designed according to the instantaneous frequency change rate of the signal;
[0012] Step 5: For multi-component signals, the time-frequency domain is divided into different regions according to the extracted ridges, and an adaptive time-frequency window is designed according to the changing trends of different components;
[0013] Step 6: construct a new short-time Fourier transform with an adaptive window function to obtain a more concentrated time-frequency coefficient;
[0014] Step 7: By detecting the local maximum of the time-frequency coefficient in the frequency direction, a new frequency redistribution operator is constructed to obtain an estimated value of the instantaneous frequency of the signal;
[0015] Step 8: Use the improved synchronization extraction operator to extract the time-frequency coefficients, remove the ambiguous time-frequency energy, and retain the signal strength information.
[0016] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the program, the improved synchronous extraction time-frequency analysis method based on adaptive window length is implemented.
[0017] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the improved synchronous extraction time-frequency analysis method based on adaptive window length.
[0018] Compared with the prior art, the present invention has the following significant advantages: 1) The present invention has an adaptive time-frequency window, which can adaptively change the window length according to the frequency modulation intensity of each component, so that the time-frequency representation of each component is clear, the time-frequency aggregation is high, and it is adaptable to various forms of signals; 2) The present invention can better extract weak amplitude signals and retain the intensity information of each component.
[0019] The present invention is further described in detail below in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] Figure 1 The flowchart of the improved synchronous extraction time-frequency analysis method based on adaptive window length of the present invention.
[0021] Figure 2 It is a diagram of the time-frequency analysis result of Example 1 of the present invention.
[0022] Figure 3 This is a diagram of the time-frequency analysis results of synchronous compression embodiment 1.
[0023] Figure 4 It is a comparison chart of the absolute error of instantaneous frequency estimation of the present invention and synchronous compression embodiment 1.
[0024] Figure 5 This is a diagram of the time-frequency analysis results of Example 2 of the present invention.
[0025] Figure 6 This is a diagram of the time-frequency analysis results of synchronous compression embodiment 2. DETAILED DESCRIPTION
[0026] Combination Figure 1 As shown, the present invention is an improved synchronous extraction time-frequency analysis method based on adaptive window length, comprising the following steps:
[0027] Step 1: Perform a preliminary short-time Fourier transform on the input signal x(t) to obtain the short-time Fourier coefficients Specifically:
[0028]
[0029] Where t is the time variable; ω is the frequency variable; τ is the integral variable; i is the complex unit; g(t) is the Gaussian window function, which is taken as Here σ is the standard deviation of the Gaussian window and represents the length of the Gaussian window;
[0030] Step 2: According to the second-order local modulation operator Estimate the frequency change rate of each time-frequency point, specifically:
[0031]
[0032] in, denote the short-time Fourier transform of window functions g′(t), g″(t), tg(t), and tg′(t) respectively;
[0033] Step 3: Solve the following optimization problem iteratively through the stripping algorithm, search for local energy peaks, and extract K time-frequency ridges of the signal:
[0034]
[0035] Among them, t n is the nth moment, n∈{1,...,N}, N is the total number of moments; l k is the kth time-frequency ridge, k∈{1,...,K}, K is the total number of signal components; l k (t n ) is the kth time-frequency ridge at t n Frequency of time; represents the short-time Fourier coefficient after stripping k-1 ridges and their neighborhoods, t is the time, and ω is the frequency variable; is the kth time-frequency ridge at t n The short-time Fourier coefficient at time ; λ is the amplification factor; is a second-order local modulation operator; γ n Yes n Custom thresholds for moments can be set The specific steps are:
[0036] Step 3-1: At time t1, search for the global energy peak along the frequency direction Determine the threshold γ1;
[0037] Step 3-2, from t1 to t N Repeat step 3-1 at each moment to obtain the threshold γ at each moment n ;
[0038] Step 3-3, at time t1, search for a local peak along the frequency direction, and the amplitude must be greater than the threshold γ1, and obtain the frequency points (ω1, ω2...) that meet the conditions;
[0039] Step 3-4: Determine the starting position of the ridge line (t1, ω1), at time t2 and the constrained frequency domain interval Search for the point with the maximum energy, and the amplitude must be greater than the threshold γ2 to obtain the time-frequency point (t2, ω2) that meets the conditions;
[0040] Step 3-5: From t1 to t N-1 Repeat steps 3-4 at all times to obtain the first time-frequency ridge line l1;
[0041] Step 3-6: Remove the ridge line l1 and its domain coefficients from the initial short-time Fourier coefficients, denoted as
[0042] Step 3-7, repeat steps 3-3 to 3-6 to obtain K time-frequency ridges;
[0043] Step 4: For a single-component signal, an adaptive time window is designed according to the instantaneous frequency change rate of the signal. The specific steps are as follows:
[0044] Step 4-1: Construct time-varying window length σ(t)
[0045]
[0046] Where l(t) is the time-frequency ridge of the signal; is the second-order local modulation operator; θ is the window length adjustment factor;
[0047] Step 4-2: Substitute σ(t) into the Gaussian window function to construct the adaptive window function g σ (t) is:
[0048]
[0049] Step 5: For multi-component signals, the time-frequency domain is divided into different regions according to the extracted ridges, and an adaptive time-frequency window is designed according to the changing trends of different components. The specific steps are as follows:
[0050] Step 5-1: Construct a window length σ that varies with time and frequency k (t, ω)
[0051] B k ≤ω≤B k+1
[0052] in, is a second-order local modulation operator; θ is the window length adjustment factor; the frequency interval of each component near its ridge point [B k , B k+1 ] The inner window length is fixed, l k (t) is the kth ridge extracted, k∈{1,...,K}, K is the total number of signal components, assuming l1(t)≤l2(t)≤…l K (t), then it is desirable
[0053]
[0054] Where M is the total number of frequency points;
[0055] Step 5-2: Set σ k Substitute (t, ω) into the Gaussian window function to construct the adaptive window function g σ(t, ω) is:
[0056]
[0057] Step 6: Construct a new short-time Fourier transform with an adaptive window function to obtain a more concentrated time-frequency coefficient F A (t, ω), specifically:
[0058]
[0059] Where t is the time variable; ω is the frequency variable; τ is the integral variable; i is the complex unit, g σ (t, ω) is the adaptive window function;
[0060] Step 7: By detecting the local maximum of the time-frequency coefficient in the frequency direction, a new frequency redistribution operator is constructed to obtain an estimate of the instantaneous frequency of the signal. Specifically:
[0061]
[0062] Among them, F A (t, ω) is the adaptive time-frequency coefficient, Δ is the defined spectral interval;
[0063] Step 8: Use the improved synchronization extraction operator ASEO to extract accurate time-frequency coefficients, remove the ambiguous time-frequency energy, and retain the signal strength information. The specific steps are as follows:
[0064] Step 8-1: Calculate the improved synchronization extraction operator based on the instantaneous frequency estimate obtained
[0065]
[0066]
[0067] Among them, η is the frequency variable; is the instantaneous frequency estimate of the signal; F A (t, ω) is the adaptive time-frequency coefficient; M is the mean of the time-frequency coefficient.
[0068] Step 8-2, and adaptive short-time Fourier coefficient F A (t, ω) cascade to obtain the adaptive improved synchronous extraction transform:
[0069]
[0070] Among them, t is the time variable; η is the frequency variable; ω is the integral variable; is the estimated value of the instantaneous frequency of the signal; F A (t, ω) is the adaptive time-frequency coefficient.
[0071] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned improved synchronous extraction time-frequency analysis method based on adaptive window length when executing the program.
[0072] And a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the improved synchronous extraction time-frequency analysis method based on adaptive window length.
[0073] The present invention is further described in detail below in conjunction with two embodiments.
[0074] Example 1
[0075] The simulation signal is the superposition of two non-linear frequency modulation signals x1(t) and x2(t), and the analytical expression is:
[0076] x1(t)=(1-0.6e -0.3t )sin(2π(38t+3sin(1.5t)))
[0077] x2(t)=e -0.4t sin(2π(15t+2sin(4.5t)))
[0078] The sampling frequency is 100Hz. Figure 2 The time-frequency spectrum obtained by the improved synchronous extraction transformation using the adaptive window length can be seen that the present invention can clearly depict the frequency changes of the two components at the same time and can reflect the intensity changes of the signal. Figure 3 The analysis results of the original synchronous compression transformation are compared. Figure 2 , Figure 3 It can be seen that the synchronous compression transform has a clearer time-frequency representation for weak frequency modulation components, but for strong frequency modulation components, the time-frequency spectrum energy diverges at the place where the frequency changes rapidly, while the time-frequency result of the present invention has a stronger aggregation. Figure 4 This is the absolute error curve of the instantaneous frequency of the stress-frequency component estimated by the present invention and the synchronous compression transformation. It can be seen that the instantaneous frequency estimation error of the present invention is smaller and more accurate.
[0079] Example 2
[0080] The simulation signal is the superposition of three non-linear frequency modulation signals x1(t), x2(t), and x3(t) with significantly different amplitudes. The analytical expression is:
[0081] x1(t)=sin(2π(38t+3sin(1.5t)))
[0082] x2(t)=0.1sin(2π(25t+3sin(1.5t)))
[0083] x3(t)=0.01sin(2π(13t+3sin(1.5t)))
[0084] The sampling frequency is 100Hz. Figure 5 The time-frequency spectrum obtained by the improved synchronous extraction transformation using the adaptive window length can be seen that the present invention can clearly represent the weak component signal. Figure 6 The analysis results of the original synchronous compression transformation are compared. Figure 5 , Figure 6 It can be seen that the synchronous compression transformation is almost unobservable for weak frequency modulation components, while the present invention can clearly characterize the time-varying trajectory of each component, has high time-frequency aggregation, and can reflect the intensity of each component.
[0085] The present invention can adapt to the frequency modulation intensity of different component signals, adaptively adjust the window length in the time and frequency directions, can clearly characterize weak signal components, can reflect the signal strength, adapt to various signal forms, finely characterize time-varying characteristics, and has high time-frequency resolution.
[0086] The above is only a preferred embodiment of the present invention. It should be pointed out that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.
Claims
1. An improved synchronous extraction time-frequency analysis method based on adaptive window length, characterized in that: The following steps are involved: Step 1: Perform a preliminary short-time Fourier transform on the input signal x(t) to obtain the short-time Fourier coefficients Specifically: Where t is the time variable; ω is the frequency variable; τ is the integral variable; i is the complex unit; g(t) is the Gaussian window function, which is taken as Here σ is the standard deviation of the Gaussian window and represents the length of the Gaussian window; Step 2: According to the second-order local modulation operator Estimate the frequency change rate of each time-frequency point, specifically: in, denote the short-time Fourier transform of window functions g′(t), g″(t), tg(t), tg′(t) respectively; Step 3, iteratively solve the optimization problem through the stripping algorithm, search for local energy peaks, extract time-frequency ridges, estimate the number of signal components, and obtain the instantaneous frequency change rate of each component; The following optimization problem is solved iteratively by the stripping algorithm to search for local energy peaks and extract K time-frequency ridges of the signal: Among them, t n is the nth moment, n∈{1,…,N}, N is the total number of moments; l k is the kth time-frequency ridge, k∈{1,…,K}, K is the total number of signal components; l k (t n ) is the kth time-frequency ridge at t n Frequency of time; represents the short-time Fourier coefficient after stripping k-1 ridges and their neighborhoods, t is the time, and ω is the frequency variable; is the kth time-frequency ridge at t n The short-time Fourier coefficient at time ; λ is the amplification factor; is a second-order local modulation operator; γ n Yes n The custom threshold at the moment is set to The specific steps are: Step 3-1: At time t1, search for the global energy peak along the frequency direction Determine the threshold γ1; Step 3-2, from t1 to t N Repeat step 3-1 at each moment to obtain the threshold γ at each moment n ; Step 3-3, at time t1, search for the local peak along the frequency direction, and the amplitude must be greater than the threshold γ1, and obtain the frequency points (ω1, ω2...) that meet the conditions; Step 3-4: Determine the starting position of the ridge line (t1, ω1), at time t2 and the constrained frequency domain interval Search for the point with the maximum energy, and the amplitude must be greater than the threshold γ2 to obtain the time-frequency point (t2, ω2) that meets the conditions; Step 3-5: From t1 to t N-1 Repeat steps 3-4 at all times to obtain the first time-frequency ridge line l1; Step 3-6: Remove the ridge line l1 and its domain coefficients from the initial short-time Fourier coefficients, denoted as Step 3-7, repeat steps 3-3 to 3-6 to obtain K time-frequency ridges; Step 4: For a single-component signal, an adaptive time window is designed according to the instantaneous frequency change rate of the signal. The specific steps are as follows: Step 4-1: Construct time-varying window length σ(t) Where l(t) is the time-frequency ridge of the signal; is the second-order local modulation operator; θ is the window length adjustment factor; Step 4-2: Substitute σ(t) into the Gaussian window function to construct the adaptive window function g σ (t) is: Step 5: For multi-component signals, the time-frequency domain is divided into different regions according to the extracted ridges, and an adaptive time-frequency window is designed according to the frequency change trends of different signal components. The specific steps are as follows: Step 5-1: Construct a window length σ that varies with time and frequency k (t,ω) in, is a second-order local modulation operator; θ is the window length adjustment factor; the frequency interval of each component near its ridge point [B k ,B k+1 ] The inner window length is fixed, l k (t) is the frequency of the kth time-frequency ridge at time t, k∈{1,…,K}, K is the total number of signal components, assuming l1(t)≤l2(t≤…l K (t), then it is desirable Where M is the total number of frequency points; Step 5-2: Set σ k Substitute (t,ω) into the Gaussian window function to construct the adaptive window function g σ (t,ω) is: Step 6: construct a new short-time Fourier transform with an adaptive window function to obtain a more concentrated time-frequency coefficient; Step 7: By detecting the local maximum of the time-frequency coefficient in the frequency direction, a new frequency redistribution operator is constructed to obtain an estimated value of the instantaneous frequency of the signal; Through the local maximum of energy, a new frequency redistribution operator is constructed to obtain an accurate estimate of the instantaneous frequency of the signal, specifically: in, is the instantaneous frequency estimate of the signal; F A (t,ω) is the adaptive time-frequency coefficient, Δ is the defined spectral interval; Step 8: Use the improved synchronization extraction operator to extract the time-frequency coefficients, remove the ambiguous time-frequency energy, and retain the signal strength information. The specific steps are as follows: Step 8-1: Calculate the improved synchronization extraction operator based on the instantaneous frequency estimate obtained Among them, η is the frequency variable; is the instantaneous frequency estimate of the signal; F A (t,ω) is the adaptive time-frequency coefficient; M is the mean of the time-frequency coefficient; Step 8-2, and adaptive short-time Fourier coefficient F A (t,ω) cascade to obtain the adaptive improved synchronous extraction transformation: Among them, t is the time variable; η is the frequency variable; ω is the integral variable.
2. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the improved synchronization extraction time-frequency analysis method based on adaptive window length as claimed in claim 1 is implemented.
3. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the improved synchronization extraction time-frequency analysis method based on adaptive window length as claimed in claim 1 is implemented.
Citation Information
Patent Citations
Self-adaptive time-frequency synchronous compression method based on Rayleigh entropy
CN107576943A
High-precision synchronous extraction generalized S transform time frequency analysis method
CN108694392A
High-order synchronous extraction transform signal time-frequency analysis method
CN109117832A
Fractional order synchronous extraction generalized S transformation time-frequency decomposition and reconstruction method
CN110347970A
Adaptive synchronous compression time-frequency analysis method based on frequency change rate
CN111639541A
