A digital signal processing method

Abstract

The present invention discloses a digital signal processing method, which includes signal acquisition, signal preprocessing, matrix transformation, signal modulation and model parameter search. The present invention belongs to the field of digital signal processing technology, and specifically refers to a digital signal processing method. The scheme performs filtering processing on the original signal through equal-interval sampling, periodic denoising processing and designing a lattice wave digital filter. The signal is transformed and decomposed through matrix changes based on orthogonal components and transformations based on fading, thereby extracting signal feature information; signal modulation is realized based on a deep learning model, and the encoder and decoder layers use Conv1D and MaxPool1D blocks to realize spatiotemporal downsampling and dimensionality reduction. The transformer layer introduces an attention mechanism to effectively capture long-range dependencies and intra-sequence correlations; the loss function considers both the reconstruction accuracy of the signal and the classification accuracy; and the optimal model parameters are searched through an improved binary search algorithm.

Description

Technical Field

[0001] The present invention relates to the technical field of digital signal processing, and in particular to a digital signal processing method. Background Art

[0002] Digital signal processing refers to the process of digitizing, filtering, spectral analysis, compression, encoding, detection, and estimation of signals using digital computers or dedicated digital signal processors. However, in general signal processing, there are problems such as serious noise interference of the original signal, interference of periodic components on signal analysis, and weak signal feature extraction capabilities; general signal processing methods have problems such as low precision and accuracy of signal modulation and weak generalization capabilities. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a digital signal processing method. In view of the problems that the original signal noise interference is serious, the periodic component interferes with the signal analysis, and the signal feature extraction ability is weak in the general signal processing process, this scheme filters the original signal through equal-interval sampling, periodic denoising and designing a lattice wave digital filter to realize the transformation and decomposition of the signal through the matrix change based on orthogonal components and the transformation based on fading, so as to extract the signal feature information; in view of the problems that the signal modulation precision and accuracy are low and the generalization ability is weak in the general signal processing method, this scheme realizes signal modulation based on a deep learning model, the encoder and decoder layers use Conv1D and MaxPool1D blocks to realize spatiotemporal downsampling and dimensionality reduction, and the Transformer layer introduces an attention mechanism to effectively capture long-range dependencies and intra-sequence correlations; the loss function considers both the reconstruction accuracy and the classification accuracy of the signal; and the optimal parameters of the model are searched through an improved binary search algorithm.

[0004] The technical solution adopted by the present invention is as follows: The present invention provides a digital signal processing method, which comprises the following steps:

[0005] Step S1: signal acquisition;

[0006] Step S2: signal preprocessing;

[0007] Step S3: matrix transformation;

[0008] Step S4: signal modulation;

[0009] Step S5: Model parameter search.

[0010] Furthermore, in step S1, the signal acquisition refers to acquiring the time domain waveform data, frequency domain information, time series data and modulation type of the signal.

[0011] Furthermore, in step S2, the signal preprocessing specifically includes the following steps:

[0012] Step S21: Sampling at equal intervals to determine the sampling frequency. The formula used is as follows:

[0013] ;

[0014] Where fm is the sampling frequency, f max is the highest frequency component of the signal to be processed;

[0015] Step S22: removing periodicity. For periodic signals, the sampled signal is subjected to periodicity removal. The formula used is as follows:

[0016] ;

[0017] In the formula, is a sequence in a discrete time system, which is the result of periodic elimination calculation; s is a complex variable of Laplace transform, which is used to convert the continuous time domain to the discrete time domain; T is the sampling period, which represents the time interval of sampling in the discrete time system;

[0018] Step S23: Design a dot wave digital filter. The filter consists of a lower branch and an upper branch. Each branch includes an all-pass component and a filter component. The function of the all-pass component is to generate a passband and a closed band. Define the parameter γ0 of the adapter as the reflection coefficient of the adapter, which represents the degree of reflection of the adapter to the signal from the all-pass component. The formula used is as follows:

[0019] ;

[0020] Where B0 is the gate resistance.

[0021] Furthermore, in step S3, the matrix transformation specifically includes the following steps:

[0022] Step S31: Perform matrix transformation on the orthogonal components of the signal, and the formula used is as follows:

[0023] ;

[0024] In the formula, is the vector form of the transformed orthogonal components, is the vector form of the original orthogonal components, is the frequency offset, t is the time, is the phase shift;

[0025] Step S32: transform the orthogonal components based on fading, as shown below:

[0026] ;

[0027] In the formula, It is a vector form of orthogonal components after fading transformation; x(t) and y(t) represent the real and imaginary parts of the signal respectively, and t is a time variable.

[0028] Furthermore, in step S4, the signal modulation specifically includes the following steps:

[0029] Step S41: Define the model architecture. The encoder layer consists of alternating Conv1D and MaxPool1D blocks. The decoder layer consists of alternating Conv1D and upsampling blocks. The Transformer layer uses the Pytorchreformer architecture with a hidden dimension of 256, a depth of 2, and 8 attention heads. The output of the Transformer layer is flattened and used as the input of a fully connected layer with an output size of 1024, followed by BatchNorm1D, ReLU, and Dropout layers. The last layer of the classifier is another fully connected layer whose output size varies depending on the current task. The dropout percentage used in the entire model is 0.2. Stochastic gradient descent is used to minimize the total loss function, while the ADAM optimizer is used to update the parameters. The learning rate is set to 0.001.

[0030] Step S42: define the L2 reconstruction loss function, the formula used is as follows:

[0031] ;

[0032] In the formula, is the L2 reconstruction loss function, which is used to measure the difference between the target signal Tx and the generated reconstructed signal The difference between is the norm; θ M are model parameters;

[0033] Step S43: define the classifier loss function, the formula used is as follows:

[0034] ;

[0035] In the formula, is the classifier loss function; CE is the cross entropy loss function; λ Ci is the loss weight; is the L2 reconstruction loss function for the modulation type Ci, i is the index of the modulation type; L is the actual modulation type; is the predictive modulation type;

[0036] Step S44: construct the model output target, which is expressed as follows:

[0037] ;

[0038] Where L is the modulation type of the target signal, is the phase shift angle; is the frequency offset of the signal; SNR is the value of the signal-to-noise ratio.

[0039] Furthermore, in step S5, the model parameter search specifically includes the following steps:

[0040] Step S51: Search initialization, determine the binary search space based on the model parameters to be searched, randomly initialize the parameter search position based on the search space, each position in the parameter search space corresponds to a set of binary codes, and use the model performance established based on the parameter search position as the fitness value of the search individual; sort the search individuals in descending order according to the fitness value, and divide the search individuals into a better group and a general group; and use the individual with the highest fitness value as the global optimal individual;

[0041] Step S52: Design a better group movement strategy, the formula used is as follows:

[0042] ;

[0043] ;

[0044] In the formula, and are the positions of the nth and n+1th moves of the I individual in the J dimension of the better group; w is the move weight; C is the better move factor, and R1 is a random number between 0 and 1; is the position of the global optimal individual J dimension at the nth iteration; sig(·) is the decision function; complement() is the complement operation;

[0045] Step S53: Design a general group mobility strategy, the formula used is as follows:

[0046] ;

[0047] ;

[0048] In the formula, is the random move weight, w max is the maximum weight, w min is the minimum weight; and is the position of the I individual in the general group in the J dimension after the n+1th and nth moves; N is the maximum number of iterations, and ε is the smallest positive value; R2 is a random number between 0 and 1, which is independent of R1;

[0049] Step S54: Search judgment. A fitness threshold is set in advance. When there is a search individual whose fitness value is higher than the search threshold, a model is established based on the individual position. When the maximum number of iterations is reached, the parameter search position is reinitialized for searching. Otherwise, the general group and the better group are re-divided, and the global optimal individual is re-selected for mobile search.

[0050] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0051] (1) In view of the problems of serious noise interference of the original signal, interference of periodic components on signal analysis, and weak signal feature extraction capability in the general signal processing process, this scheme filters the original signal through equal-interval sampling, periodic denoising and design of lattice wave digital filter. The signal is transformed and decomposed through matrix changes based on orthogonal components and transformation based on fading, thereby extracting signal feature information.

[0052] (2) In view of the problems of low precision and accuracy of signal modulation and weak generalization ability in general signal processing methods, this scheme implements signal modulation based on a deep learning model. The encoder and decoder layers use Conv1D and MaxPool1D blocks to achieve spatiotemporal downsampling and dimensionality reduction. The Transformer layer introduces an attention mechanism to effectively capture long-range dependencies and intra-sequence correlations. The loss function considers both the reconstruction accuracy and the classification accuracy of the signal. The optimal model parameters are searched through an improved binary search algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 A schematic flow chart of a digital signal processing method provided by the present invention;

[0054] Figure 2 is a schematic flow chart of step S2;

[0055] Figure 3 is a schematic flow chart of step S4;

[0056] Figure 4 It is a flowchart diagram of step S5.

[0057] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0058] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0059] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0060] Example 1, see Figure 1 The present invention provides a digital signal processing method, which comprises the following steps:

[0061] Step S1: signal acquisition, collecting the signal's time domain waveform data, frequency domain information, time series data and modulation type;

[0062] Step S2: signal preprocessing, completing signal filtering processing through equal-interval sampling, periodic removal operation and designing a lattice wave digital filter;

[0063] Step S3: Matrix transformation, capturing the signal structure and characteristics by performing matrix transformation on the orthogonal components of the signal and transforming the orthogonal components based on fading;

[0064] Step S4: Signal modulation. The signal is modulated by designing a deep learning model. The encoder and decoder layers use Conv1D and MaxPool1D blocks to achieve spatiotemporal downsampling and dimensionality reduction, and the Transformer layer introduces an attention mechanism.

[0065] Step S5: Model parameter search, which is achieved by designing a binary search strategy.

[0066] Example 2, see Figure 1 and Figure 2 This embodiment is based on the above embodiment, and the signal preprocessing specifically includes the following steps:

[0067] Step S21: Sampling at equal intervals to determine the sampling frequency. The formula used is as follows:

[0068] ;

[0069] Where fm is the sampling frequency, f maxis the highest frequency component of the signal to be processed;

[0070] Step S22: removing periodicity. For periodic signals, the sampled signal is subjected to periodicity removal. The formula used is as follows:

[0071] ;

[0072] In the formula, is a sequence in a discrete time system, which is the result of periodic elimination calculation; s is a complex variable of Laplace transform, which is used to convert the continuous time domain to the discrete time domain; T is the sampling period, which represents the time interval of sampling in the discrete time system;

[0073] Step S23: Design a dot wave digital filter. The filter consists of a lower branch and an upper branch. Each branch includes an all-pass component and a filter component. The function of the all-pass component is to generate a passband and a closed band. Define the parameter γ0 of the adapter as the reflection coefficient of the adapter, which represents the degree of reflection of the adapter to the signal from the all-pass component. The formula used is as follows:

[0074] ;

[0075] Where B0 is the gate resistance.

[0076] Example 3, see Figure 1 This embodiment is based on the above embodiment. In step S3, the matrix transformation specifically includes the following steps:

[0077] Step S31: Perform matrix transformation on the orthogonal components of the signal, and the formula used is as follows:

[0078] ;

[0079] In the formula, is the vector form of the transformed orthogonal components, is the vector form of the original orthogonal components, is the frequency offset, t is the time, is the phase shift;

[0080] Step S32: transform the orthogonal components based on fading, as shown below:

[0081] ;

[0082] In the formula, It is a vector form of orthogonal components after fading transformation; x(t) and y(t) represent the real and imaginary parts of the signal respectively, and t is a time variable.

[0083] By performing the above operations, in order to address the problems of serious noise interference of the original signal, interference of periodic components on signal analysis, and weak signal feature extraction capability in the general signal processing process, this scheme filters the original signal through equal-interval sampling, periodic denoising and design of lattice wave digital filters, and realizes transform decomposition of the signal through matrix changes based on orthogonal components and transformation based on fading, thereby extracting signal feature information.

[0084] Example 4, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. In step S4, the signal modulation specifically includes the following steps:

[0085] Step S41: Define the model architecture. The encoder layer consists of alternating Conv1D and MaxPool1D blocks. The decoder layer consists of alternating Conv1D and upsampling blocks. The Transformer layer uses the Pytorchreformer architecture with a hidden dimension of 256, a depth of 2, and 8 attention heads. The output of the Transformer layer is flattened and used as the input of a fully connected layer with an output size of 1024, followed by BatchNorm1D, ReLU, and Dropout layers. The last layer of the classifier is another fully connected layer whose output size varies depending on the current task. The dropout percentage used in the entire model is 0.2. Stochastic gradient descent is used to minimize the total loss function, while the ADAM optimizer is used to update the parameters. The learning rate is set to 0.001.

[0086] Step S42: define the L2 reconstruction loss function, the formula used is as follows:

[0087] ;

[0088] In the formula, is the L2 reconstruction loss function, which is used to measure the difference between the target signal Tx and the generated reconstructed signal The difference between is the norm; θ M are model parameters;

[0089] Step S43: define the classifier loss function, the formula used is as follows:

[0090] ;

[0091] In the formula, is the classifier loss function; CE is the cross entropy loss function; λ Ci is the loss weight; is the L2 reconstruction loss function for the modulation type Ci, i is the index of the modulation type; L is the actual modulation type; is the predictive modulation type;

[0092] Step S44: construct the model output target, which is expressed as follows:

[0093] ;

[0094] Where L is the modulation type of the target signal, is the phase shift angle; is the frequency offset of the signal; SNR is the value of the signal-to-noise ratio.

[0095] Example 5, see Figure 1 and Figure 4 This embodiment is based on the above embodiment. In step S5, the model parameter search specifically includes the following steps:

[0096] Step S51: Search initialization, determine the binary search space based on the model parameters to be searched, randomly initialize the parameter search position based on the search space, each position in the parameter search space corresponds to a set of binary codes, and use the model performance established based on the parameter search position as the fitness value of the search individual; sort the search individuals in descending order according to the fitness value, and divide the search individuals into a better group and a general group; and use the individual with the highest fitness value as the global optimal individual;

[0097] Step S52: Design a better group movement strategy, the formula used is as follows:

[0098] ;

[0099] ;

[0100] In the formula, and are the positions of the nth and n+1th moves of the I individual in the J dimension of the better group; w is the move weight; C is the better move factor, and R1 is a random number between 0 and 1; is the position of the global optimal individual J dimension at the nth iteration; sig(·) is the decision function; complement() is the complement operation;

[0101] Step S53: Design a general group mobility strategy, the formula used is as follows:

[0102] ;

[0103] ;

[0104] In the formula, is the random move weight, w max is the maximum weight, w min is the minimum weight; and is the position of the I individual in the general group in the J dimension after the n+1th and nth moves; N is the maximum number of iterations, and ε is the smallest positive value; R2 is a random number between 0 and 1, which is independent of R1;

[0105] Step S54: Search judgment. A fitness threshold is set in advance. When there is a search individual whose fitness value is higher than the search threshold, a model is established based on the individual position. When the maximum number of iterations is reached, the parameter search position is reinitialized for searching. Otherwise, the general group and the better group are re-divided, and the global optimal individual is re-selected for mobile search.

[0106] By performing the above operations, in view of the problems of low precision and accuracy of signal modulation and weak generalization ability in general signal processing methods, this scheme realizes signal modulation based on deep learning model. Conv1D and MaxPool1D blocks are used in encoder and decoder layers to realize spatiotemporal downsampling and dimensionality reduction. The attention mechanism is introduced in Transformer layer to effectively capture long-range dependencies and intra-sequence correlations. The loss function considers both the reconstruction accuracy and classification accuracy of the signal. And the optimal parameters of the model are searched through the improved binary search algorithm.

[0107] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0108] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

[0109] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A digital signal processing method, characterized in that: The method comprises the following steps: Step S1: signal acquisition; Step S2: signal preprocessing; Step S3: matrix transformation; Step S4: signal modulation; Step S5: model parameter search; In step S5, the model parameter search specifically includes the following steps: Step S51: Search initialization, determine the binary search space based on the model parameters to be searched, randomly initialize the parameter search position based on the search space, each position in the parameter search space corresponds to a set of binary codes, and use the model performance established based on the parameter search position as the fitness value of the search individual; sort the search individuals in descending order according to the fitness value, and divide the search individuals into a better group and a general group; and use the individual with the highest fitness value as the global optimal individual; Step S52: Design a better group movement strategy, the formula used is as follows: ； ； In the formula, and are the positions of the nth and n+1th moves of the I individual in the J dimension of the better group; w is the move weight; C is the better move factor, and R1 is a random number between 0 and 1; is the position of the global optimal individual J dimension at the nth iteration; sig(·) is the decision function; complement() is the complement operation; Step S53: Design a general group mobility strategy, the formula used is as follows: ； ； In the formula, is the random move weight, w max is the maximum weight, w min is the minimum weight; and is the position of the I individual in the general group in the J dimension after the n+1th and nth moves; N is the maximum number of iterations, and ε is the smallest positive value; R2 is a random number between 0 and 1, which is independent of R1; Step S54: Search determination, a fitness threshold is set in advance, when there is a search individual whose fitness value is higher than the search threshold, a model is established based on the individual position; when the maximum number of iterations is reached, the parameter search position is reinitialized for searching; otherwise, the general group and the better group are re-divided, and the global optimal individual is re-selected for mobile search; In step S4, the signal modulation specifically includes the following steps: Step S41: Define the model architecture. The encoder layer consists of alternating Conv1D and MaxPool1D blocks. The decoder layer consists of alternating Conv1D and upsampling blocks. The Transformer layer uses the Pytorchreformer architecture with a hidden dimension of 256, a depth of 2, and 8 attention heads. The output of the Transformer layer is flattened and used as the input of a fully connected layer with an output size of 1024, followed by BatchNorm1D, ReLU, and Dropout layers. The last layer of the classifier is another fully connected layer whose output size varies depending on the current task. The dropout percentage used in the entire model is 0.

2. Stochastic gradient descent is used to minimize the total loss function, while the ADAM optimizer is used to update the parameters. The learning rate is set to 0.

001. Step S42: define the L2 reconstruction loss function, the formula used is as follows: ； In the formula, is the L2 reconstruction loss function, which is used to measure the difference between the target signal Tx and the generated reconstructed signal The difference between is the norm; θ M are model parameters; Step S43: define the classifier loss function, the formula used is as follows: ； In the formula, is the classifier loss function; CE is the cross entropy loss function; λ Ci is the loss weight; is the L2 reconstruction loss function for the modulation type Ci, i is the index of the modulation type; L is the actual modulation type; is the predictive modulation type; Step S44: construct the model output target, which is expressed as follows: ； Where L is the modulation type of the target signal, is the phase shift angle; is the frequency offset of the signal; SNR is the value of the signal-to-noise ratio.

2. A digital signal processing method according to claim 1, characterized in that: In step S2, the signal preprocessing specifically includes the following steps: Step S21: Sampling at equal intervals to determine the sampling frequency. The formula used is as follows: ； Where fm is the sampling frequency, f max is the highest frequency component of the signal to be processed; Step S22: removing periodicity. For periodic signals, the sampled signal is subjected to periodicity removal. The formula used is as follows: ； In the formula, is a sequence in a discrete time system, which is the result of periodic elimination calculation; s is the complex variable of Laplace transform, which is used to convert the continuous time domain to the discrete time domain; T is the sampling period, which represents the time interval of sampling in the discrete time system; Step S23: Design a dot wave digital filter. The filter consists of a lower branch and an upper branch. Each branch includes an all-pass component and a filter component. The function of the all-pass component is to generate a passband and a closed band. Define the parameter γ0 of the adapter as the reflection coefficient of the adapter, which represents the degree of reflection of the adapter to the signal from the all-pass component. The formula used is as follows: ； Where B0 is the gate resistance.

3. A digital signal processing method according to claim 1, characterized in that: In step S3, the matrix transformation specifically includes the following steps: Step S31: Perform matrix transformation on the orthogonal components of the signal, and the formula used is as follows: ； In the formula, is the vector form of the transformed orthogonal components, is the vector form of the original orthogonal components, is the frequency offset, t is the time, is the phase shift; Step S32: transform the orthogonal components based on fading, as shown below: ； In the formula, It is a vector form of orthogonal components after fading transformation; x(t) and y(t) represent the real and imaginary parts of the signal respectively, and t is a time variable.

4. A digital signal processing method according to claim 1, characterized in that: In step S1, the signal acquisition refers to acquiring the time domain waveform data, frequency domain information, time series data and modulation type of the signal.

Images