A method, apparatus, device, and medium for separating mixed signals

Abstract

This invention discloses a method, apparatus, device, and medium for separating mixed signals. The method includes: acquiring the weights and encoding results of a mixed signal; performing feature aggregation on the encoding results to obtain aggregated features, and determining at least one similarity matrix between the aggregated features and the encoding results based on the aggregated features and the encoding results; weighting each of the similarity matrices using the weights to obtain at least one separation matrix; fusing the encoding results and each of the separation matrices, and decoding to obtain a separated clean signal. The embodiments of this invention can improve the mixed signal separation effect, reduce the dependence on signal type when separating mixed signals, and improve the versatility of the mixed signal separation method.

Description

Technical Field

[0001] This invention relates to the field of signal processing technology, and in particular to a method, apparatus, device, and medium for separating mixed signals. Background Technology

[0002] With the rapid development and widespread application of wireless communication technology, especially with the advent of the Internet of Things era, spectrum resources are becoming increasingly scarce and the electromagnetic environment is becoming increasingly complex. Particularly in non-cooperative reception scenarios, it is becoming increasingly common for a single antenna to receive other signals, resulting in signal interference that affects communication quality. Signal processing during communication generally requires high signal-to-noise ratio signals, and how to separate mixed signals is a research direction in wireless communication technology.

[0003] For mixed signal separation, general blind signal separation methods based on signal characteristics are heavily dependent on signal type. The proposed methods are often only applicable to specific types of signals, and the established algorithm models are difficult to design and have poor versatility. Summary of the Invention

[0004] This invention provides a mixed signal separation method, apparatus, device, and medium to improve the mixed signal separation effect, reduce the dependence on signal type when separating mixed signals, and improve the versatility of the mixed signal separation method.

[0005] According to one aspect of the present invention, a mixed signal separation method is provided, the method comprising:

[0006] Obtain the weights and encoding results of the mixed signal;

[0007] The encoding result is subjected to feature aggregation to obtain aggregated features, and at least one similarity matrix between the aggregated features and the encoding result is determined based on the aggregated features and the encoding result;

[0008] By weighting each of the similarity matrices using the aforementioned weights, at least one separation matrix is ​​obtained;

[0009] The encoding result and each of the separation matrices are fused together and then decoded to obtain the separated pure signal.

[0010] According to another aspect of the present invention, a mixed signal separation device is provided, the device comprising:

[0011] The information acquisition module is used to acquire the weights and encoding results of the mixed signal;

[0012] The similarity matrix calculation module is used to perform feature aggregation on the encoding result to obtain aggregated features, and determine at least one similarity matrix between the aggregated features and the encoding result based on the aggregated features and the encoding result;

[0013] A separation matrix calculation module is used to weight each of the similarity matrices using the weights to obtain at least one separation matrix;

[0014] The signal separation module is used to fuse the encoding result and each of the separation matrices, and then decode them to obtain the separated clean signal.

[0015] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0016] At least one processor; and

[0017] A memory communicatively connected to the at least one processor; wherein,

[0018] The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the mixed signal separation method according to any embodiment of the present invention.

[0019] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement the mixed-signal separation method according to any embodiment of the present invention.

[0020] According to another aspect of the present invention, a computer program product is provided, the computer program product comprising a computer program that, when executed by a processor, implements the mixed signal separation method according to any embodiment of the present invention.

[0021] The technical solution of this invention obtains the weights and encoding results of the mixed signal, performs feature aggregation on the encoding results to obtain aggregated features, determines the similarity matrix between the aggregated features and the encoding results based on the aggregated features and the encoding results, and achieves the fusion of aggregated features and encoding results. At least one separation matrix is ​​obtained by weighting the similarity matrix with weights. After fusing the encoding results and the separation matrix, the fused result is decoded to obtain a clean signal. Different weights are obtained for different types of mixed signals. Considering the weights of the mixed signals during the process of obtaining the separation matrix enables the separation of different types of mixed signals, reducing the dependence on signal type when separating mixed signals, thereby improving the versatility of the mixed signal separation method. Simultaneously, the calculation of the separation matrix considers the influence of the encoding results, aggregated features, and weights on the calculation of the separation matrix, which can improve the accuracy of the separation matrix calculation and thus improve the separation effect of the mixed signal.

[0022] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart of a mixed signal separation method provided in Embodiment 1 of the present invention;

[0025] Figure 2 This is a flowchart of a mixed signal separation method provided in Embodiment 2 of the present invention;

[0026] Figure 3 This is a flowchart of a mixed signal separation method provided in Embodiment 3 of the present invention;

[0027] Figure 4 This is a flowchart of a mixed signal separation method provided in Embodiment 4 of the present invention;

[0028] Figure 5a This is a system structure diagram of a mixed signal separation method provided in Embodiment 5 of the present invention;

[0029] Figure 5b This is a structural diagram of a classification matrix calculator provided in Embodiment 5 of the present invention;

[0030] Figure 6 This is a schematic diagram of a mixed signal separation device according to Embodiment Six of the present invention;

[0031] Figure 7 This is a schematic diagram of the structure of an electronic device that implements the mixed signal separation method of the present invention. Detailed Implementation

[0032] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0033] It should be noted that the term "comprising" and any variations thereof in the specification, claims and accompanying drawings of this invention are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such process, method, product or device.

[0034] Example 1

[0035] Figure 1 The flowchart illustrates a mixed-signal separation method according to Embodiment 1 of the present invention. This embodiment is applicable to the separation of different types of mixed signals. The method can be executed by a mixed-signal separation device, which can be implemented in hardware and / or software and can be configured in an electronic device. Figure 1 As shown, the method includes:

[0036] S110, Obtain the weights and encoding results of the mixed signal.

[0037] A mixed signal refers to a signal that needs to be separated. It can also be understood as a mixed signal composed of at least two superimposed signals. Weights are used to describe the distribution of signal characteristics in the mixed signal. Signal characteristics can be at least one of the following: energy distribution, time-domain characteristics, and frequency-domain characteristics. For example, extracting the energy distribution characteristics of the mixed signal; the denser the energy distribution, the smaller the corresponding weight. Encoding refers to the process of converting a signal from one form or format to another. The encoding result refers to the result obtained after encoding the mixed signal. In this application, the mixed signal can be encoded using an encoder to obtain the encoding result. The encoding result can be described using a time series.

[0038] Specifically, the weights of the mixed signal are obtained by performing signal feature analysis on the mixed signal, and the encoding result of the mixed signal is obtained by encoding the mixed signal.

[0039] S120. Perform feature aggregation on the encoding result to obtain aggregated features, and determine at least one similarity matrix between the aggregated features and the encoding result based on the aggregated features and the encoding result.

[0040] Feature aggregation refers to the extraction of features from the encoded result. Specifically, features can be aggregated from the encoded result using at least one deep learning network. For example, the deep learning network can be a TCN (Temporal Convolutional Network), an RNN (Recursive Neural Network), or an LSTM (Long Short-Term Memory) network. Aggregated features refer to the result after feature aggregation of the encoded result. Aggregating the encoded result using at least one deep learning network can gather the features of the encoded result, providing a basis for calculating the separation matrix.

[0041] A similarity matrix indicates the degree of similarity between the encoded result and the aggregated features. It is calculated to fuse the aggregated features and the encoded result. At least one similarity matrix is ​​required. Understandably, the similarity matrix calculated between the encoded result and the aggregated features will differ across different dimensions.

[0042] Specifically, aggregated features are obtained by aggregating the encoded results through at least one deep learning network. Then, the similarity between the aggregated features and the encoded results is calculated to determine at least one similarity matrix between the aggregated features and the encoded results.

[0043] Optionally, similarity can be calculated directly based on the aggregated features and encoding results, or the aggregated features and encoding results can be passed through the same number of convolutional layers, and then the similarity can be calculated based on the convolution results. For example, the convolutional layers can be one-dimensional convolutions. By performing multiple convolutions on the aggregated features and encoding results, feature extraction can be performed on the aggregated features and encoding results, thereby improving the accuracy of the similarity matrix calculation.

[0044] S130. The similarity matrices are weighted by the weights to obtain at least one separation matrix.

[0045] A separation matrix is ​​used to separate the encoded results. It can also be understood as decomposing the encoded result into different codes using a separation matrix. At least one separation matrix is ​​required. A single code can be extracted from the encoded result using a single separation matrix. This can be understood as follows: aliasing multiple original signals creates a mixed signal; encoding each original signal separately and then aliasing the resulting codes results in the mixed signal yields the same encoded result. A separation matrix can decompose the encoded result of the mixed signal into the codes of the original signals. Weights are applied to the similarity matrices, which can be achieved by multiplying the weights by the individual similarity matrices.

[0046] Specifically, the weights are multiplied by each similarity matrix to achieve weighted similarity matrix calculation. Based on the weighted similarity matrices, at least one separation matrix is ​​determined. For example, the weighted similarity matrices can be directly determined as the separation matrix, or the weighted similarity matrices can be further processed to determine the separation matrix. The further processing can be summing and averaging.

[0047] S140. The encoding result and each of the separation matrices are fused together and decoded to obtain the separated pure signal.

[0048] Decoding is a process of restoring the encoded result to its represented signal using a specific method. Decoding corresponds to the encoding process. In this application, decoding can be performed using a decoder. The clean signal refers to the signal before it is aliased into a mixed signal, also known as the source signal. The number of clean signals is at least two. Fusing the encoded result and each separation matrix can be achieved by multiplying the encoded result with each separation matrix by a dot product.

[0049] Specifically, the encoding result is fused with each separation matrix to obtain the fused encoding result, which is the encoding result of the pure signal. The decoder then decodes the fused encoding result to obtain the separated pure signal.

[0050] The technical solution of this invention obtains the weights and encoding results of the mixed signal, performs feature aggregation on the encoding results to obtain aggregated features, determines the similarity matrix between the aggregated features and the encoding results based on the aggregated features and the encoding results, and achieves the fusion of aggregated features and encoding results. At least one separation matrix is ​​obtained by weighting the similarity matrix with weights. After fusing the encoding results and the separation matrix, the fused result is decoded to obtain a clean signal. Different weights are obtained for different types of mixed signals. Considering the weights of the mixed signals during the process of obtaining the separation matrix enables the separation of different types of mixed signals, reducing the dependence on signal type when separating mixed signals, thereby improving the versatility of the mixed signal separation method. Simultaneously, the calculation of the separation matrix considers the influence of the encoding results, aggregated features, and weights on the calculation of the separation matrix, which can improve the accuracy of the separation matrix calculation and thus improve the separation effect of the mixed signal.

[0051] Example 2

[0052] Figure 2This is a flowchart of a mixed signal separation method provided in Embodiment 2 of the present invention. Based on the above embodiments, this embodiment specifies the step of determining at least one similarity matrix between the aggregated features and the encoding results as follows: similarity calculation is performed on the aggregated features and the encoding results in at least two dimensions to obtain at least two similarity matrices. For example... Figure 2 As shown, the method includes:

[0053] S210. Obtain the weights and encoding results of the mixed signal.

[0054] S220. Perform feature aggregation on the encoding result to obtain aggregated features, and calculate the similarity between the aggregated features and the encoding result in at least two dimensions to obtain at least two similarity matrices.

[0055] Dimensions are used to differentiate the descriptive results of data. The same data can have different descriptive results on different dimensions. Specifically, aggregated features are obtained by aggregating the encoded results through at least one convolutional layer. Dimensional transformation is then performed on the aggregated features and / or encoded results to obtain aggregated feature data and encoded result data in at least two dimensions, ensuring that the dimensions of the aggregated features and encoded results are the same. Based on the aggregated feature data and encoded result data in different dimensions, a similarity matrix is ​​calculated between the aggregated features and encoded results. At least one similarity matrix can be calculated for each dimension. For example, the dimension can be the time domain. After transforming the aggregated features and encoded results to the time domain respectively, the similarity between the aggregated features and encoded results is calculated in the time domain to obtain at least one similarity matrix.

[0056] S230. The similarity matrices are weighted by the weights to obtain at least one separation matrix.

[0057] S240. The encoding result and each of the separation matrices are fused together and decoded to obtain the separated pure signal.

[0058] The technical solution of this invention calculates the similarity between aggregated features and encoding results in at least two dimensions, thereby obtaining a similarity matrix in at least two dimensions. Based on the similarity matrices in different dimensions, at least one separation matrix is ​​obtained, so that each separation matrix can have features in different dimensions, thereby improving the separation effect of mixed signals.

[0059] Based on the above embodiments, the dimension includes the time domain and the frequency domain, and the similarity matrix includes: at least one initial time-domain similarity matrix and at least one initial frequency-domain similarity matrix, wherein the number of initial time-domain similarity matrices is the same as the number of initial frequency-domain similarity matrices; the step of weighting each of the similarity matrices to obtain at least one separation matrix includes: grouping the similarity matrices to obtain at least one similarity matrix group, wherein a similarity matrix group includes an initial time-domain similarity matrix and an initial frequency-domain similarity matrix; for each similarity matrix group, performing feature extraction on the initial time-domain similarity matrix and the initial frequency-domain similarity matrix of the similarity matrix group to obtain candidate time-domain similarity matrices and candidate frequency-domain similarity matrices of the similarity matrix group; weighting the candidate time-domain similarity matrix and the candidate frequency-domain similarity matrix of the similarity matrix group according to the weight to determine the fusion result of the similarity matrix group; and determining at least one separation matrix according to the fusion result of each of the similarity matrix groups, wherein the number of separation matrices is the same as the number of clean signals included in the mixed signal.

[0060] The initial temporal similarity matrix refers to the similarity matrix calculated in the time domain based on aggregated features and coding results. The initial frequency domain similarity matrix refers to the similarity matrix calculated in the frequency domain based on aggregated features and coding results.

[0061] Specifically, similarity calculations are performed on the aggregated features and encoding results in both the time and frequency domains. At least one initial time-domain similarity matrix is ​​obtained in the time domain, and at least one initial frequency-domain similarity matrix is ​​obtained in the frequency domain. The number of initial time-domain similarity matrices and initial frequency-domain similarity matrices are the same.

[0062] A similarity matrix group is the result of grouping at least one similarity matrix. Within a similarity matrix group, the dimensions of the similarity matrices differ. Each similarity matrix group contains the same number of similarity matrices. Specifically, when the dimensions include both the time domain and the frequency domain, an initial time-domain similarity matrix and an initial frequency-domain similarity matrix are grouped together to form a similarity matrix group.

[0063] The candidate temporal similarity matrix refers to the similarity matrix obtained after feature extraction from the initial temporal similarity matrix. The candidate frequency domain similarity matrix refers to the similarity matrix obtained after feature extraction from the initial frequency domain similarity matrix. Specifically, features can be extracted from both the initial temporal and frequency domain similarity matrices using the same number of convolutional layers.

[0064] The fusion result refers to the fused matrix obtained by weighting and fusing the candidate time-domain similarity matrix and the candidate frequency-domain similarity matrix in the same similarity matrix group. Specifically, the candidate time-domain similarity matrix and the candidate frequency-domain similarity matrix can be converted to the same dimension and updated. The updated candidate time-domain similarity matrix and the candidate frequency-domain similarity matrix are then weighted, and the weighted candidate time-domain similarity matrix and the candidate frequency-domain similarity matrix are added together to achieve fusion and obtain the fusion result. Alternatively, the candidate time-domain similarity matrix and the candidate frequency-domain similarity matrix can be weighted, converted to the same dimension, and added together to achieve fusion and obtain the fusion result. Another option is to convert the initial time-domain similarity matrix and the initial frequency-domain similarity matrix to the same dimension, perform feature extraction to obtain candidate time-domain similarity matrices and candidate frequency-domain similarity matrices, and then directly weight the candidate time-domain similarity matrix and the candidate frequency-domain similarity matrix using weights before adding them together to achieve fusion and obtain the fusion result. The conversion to the same dimension can be either converting the candidate frequency domain similarity matrix in the frequency domain to the time domain, converting the candidate time domain similarity matrix in the time domain to the frequency domain, or converting both the candidate time domain similarity matrix and the candidate frequency domain similarity matrix to other dimensions simultaneously.

[0065] Specifically, the similarity matrices are grouped according to their dimensions to obtain at least one similarity matrix group. Each similarity matrix group includes an initial time-domain similarity matrix and an initial frequency-domain similarity matrix. Within each similarity matrix group, features are extracted from both the initial time-domain and initial frequency-domain similarity matrices using the same number of convolutional layers. The feature extraction results of the initial time-domain similarity matrix are used as candidate time-domain similarity matrices, and the feature extraction results of the initial frequency-domain similarity matrix are used as candidate frequency-domain similarity matrices. The candidate time-domain and candidate frequency-domain similarity matrices in the similarity matrix group are weighted using the weights of the mixed signal, and then fused to obtain the fused result of the similarity matrix group. For the fused results obtained from each similarity matrix group, a separation matrix is ​​determined, where the number of fused results is an integer multiple of the number of separation matrices. For example, when the number of fusion results is the same as the number of separation matrices, the fusion results can be directly determined as separation matrices; when the number of fusion results is *a* times the number of separation matrices, the *a* fusion results are grouped together, the fusion results within each group are summed and averaged, and the calculated result is determined as the separation matrix. The number of separation matrices is the same as the number of pure signals contained in the mixed signal.

[0066] Similarity matrices are calculated in both the time and frequency domains. These matrices are then grouped according to their dimensions to obtain similarity matrix groups. Within these groups, features are extracted from the initial time-domain and frequency-domain similarity matrices. Then, the candidate time-domain and candidate frequency-domain similarity matrices are weighted according to their respective weights to determine the fusion result of the similarity matrix groups. The separation matrix is ​​then determined based on this fusion result. The fusion result incorporates features of the similarity matrices in both the time and frequency domains, which enhances the influence of these features on the separation matrix, increases the differences between the separation matrices, and thus improves the separation effect of the mixed signal.

[0067] Example 3

[0068] Figure 3 This is a flowchart of a mixed signal separation method provided in Embodiment 3 of the present invention. Based on the above embodiments, this embodiment extracts features from the initial temporal similarity matrix and initial frequency similarity matrix of the similarity matrix group to obtain candidate temporal similarity matrices and candidate frequency similarity matrices of the similarity matrix group. Specifically, this involves: convolving the initial temporal similarity matrix through the first convolutional layer in a series of concatenated convolutional layers to obtain the intermediate temporal similarity matrix of the first convolutional layer; convolving the initial frequency similarity matrix through the first convolutional layer to obtain the intermediate frequency similarity matrix of the first convolutional layer; and obtaining the previous convolutional matrix through the other convolutional layers besides the first convolutional layer. The intermediate temporal similarity matrix of the convolutional layer, the intermediate frequency-domain similarity matrix of the previous convolutional layer, and the intermediate temporal similarity matrices of other groups in the previous convolutional layer are used to determine the intermediate temporal similarity matrix of the other convolutional layers. The intermediate frequency-domain similarity matrix of the previous convolutional layer, the intermediate frequency-domain similarity matrix of the previous convolutional layer, and the intermediate frequency-domain similarity matrices of other groups in the previous convolutional layer are obtained through the convolutional layers other than the first one, and the intermediate frequency-domain similarity matrix of the other convolutional layers is determined. The intermediate temporal similarity matrix of the last convolutional layer in the concatenated multiple convolutional layers is determined as the candidate temporal similarity matrix of the similarity matrix group, and the intermediate frequency-domain similarity matrix of the last convolutional layer is determined as the candidate frequency-domain similarity matrix of the similarity matrix group. Figure 3 As shown, the method includes:

[0069] S301. Obtain the weights and encoding results of the mixed signal.

[0070] S302. The encoding result is subjected to feature aggregation to obtain aggregated features, and the similarity between the aggregated features and the encoding result is calculated in at least two dimensions to obtain at least two similarity matrices; the dimensions include the time domain and the frequency domain, and the similarity matrix includes: at least one initial time domain similarity matrix and at least one initial frequency domain similarity matrix, wherein the number of initial time domain similarity matrices is the same as the number of initial frequency domain similarity matrices.

[0071] S303. The similarity matrix is ​​grouped to obtain at least one similarity matrix group, wherein a similarity matrix group includes an initial time-domain similarity matrix and an initial frequency-domain similarity matrix.

[0072] S304. The initial temporal similarity matrix is ​​convolved by the first convolutional layer among multiple convolutional layers to obtain the intermediate temporal similarity matrix of the first convolutional layer.

[0073] Concatenated convolutional layers refer to concatenated convolutional layers where the initial temporal similarity matrix is ​​sequentially passed through multiple convolutional layers. This can also be understood as the output of one convolutional layer serving as the input to the next. For example, each convolutional layer can be a one-dimensional convolution. Each similarity matrix group corresponds to an initial frequency domain similarity matrix within a concatenated convolutional layer. The intermediate temporal similarity matrix refers to the output of each concatenated convolutional layer corresponding to the initial temporal similarity matrix. The intermediate temporal similarity matrix of the first convolutional layer is the convolution result obtained after inputting the initial temporal similarity matrix into the first concatenated convolutional layer.

[0074] Specifically, the initial temporal similarity matrix is ​​input into the first convolutional layer among multiple convolutional layers. The first convolutional layer convolves the initial temporal similarity matrix, and the convolution result is the intermediate temporal similarity matrix of the first convolutional layer, which is the output result of the first convolutional layer.

[0075] Understandably, before inputting the initial temporal similarity matrix and the initial frequency similarity matrix into multiple concatenated convolutional layers, these matrices can be transformed into matrices of the same dimension. For example, the initial temporal similarity matrix can be transformed into the frequency domain, or vice versa, ensuring consistent matrix dimensions and providing a foundation for subsequent calculations.

[0076] The initial frequency domain similarity matrix and the initial frequency domain similarity matrix in the similarity matrix group are input in parallel to the corresponding cascaded convolutional layers and calculated simultaneously. The similarity matrix groups are input in parallel and calculated simultaneously across different similarity matrix groups.

[0077] S305. The initial frequency domain similarity matrix is ​​convolved through the first convolutional layer to obtain the intermediate frequency domain similarity matrix of the first convolutional layer.

[0078] The intermediate frequency domain similarity matrix refers to the output of each convolutional layer in a series of concatenated convolutional layers corresponding to the initial frequency domain similarity matrix. The intermediate frequency domain similarity matrix of the first convolutional layer refers to the convolution result obtained after the initial frequency domain similarity matrix is ​​input into the first convolutional layer in a series of concatenated convolutional layers.

[0079] Specifically, the initial frequency domain similarity matrix is ​​input into the first convolutional layer among multiple convolutional layers connected in series. The first convolutional layer convolves the initial frequency domain similarity matrix, and the convolution result is the intermediate frequency domain similarity matrix of the first convolutional layer, which is the output result of the first convolutional layer.

[0080] S306. Using the concatenated convolutional layers other than the first convolutional layer, obtain the intermediate temporal similarity matrix of the previous convolutional layer, the intermediate frequency similarity matrix of the previous convolutional layer, and the intermediate temporal similarity matrix of the other groups of the previous convolutional layer, and determine the intermediate temporal similarity matrix of the other convolutional layers.

[0081] Other convolutional layers refer to all convolutional layers other than the first one in a series of concatenated convolutional layers. For example, if there are three concatenated convolutional layers: the first, second, and third, and the first convolutional layer is the first convolutional layer, then the other convolutional layers are either the second or third convolutional layers. The preceding convolutional layer refers to the convolutional layer preceding the other convolutional layers. The intermediate temporal similarity matrix of the preceding convolutional layer refers to the intermediate temporal similarity matrix output by the preceding convolutional layer for the current similarity matrix group. The intermediate frequency domain similarity matrix of the preceding convolutional layer refers to the intermediate frequency domain similarity matrix output by the preceding convolutional layer for the current similarity matrix group. Other groups refer to the similarity matrix groups other than the current similarity matrix group. The intermediate temporal similarity matrix of other groups refers to the intermediate temporal similarity matrix corresponding to the similarity matrix group other than the current similarity matrix group. The intermediate temporal similarity matrix of other groups in the previous convolutional layer refers to the intermediate temporal similarity matrix output by the previous convolutional layer for other similarity matrix groups. For example, the similarity matrix groups include group 1, group 2, and group 3. During the calculation of group 2, when calculating the intermediate temporal similarity matrix of the second convolutional layer, the intermediate temporal similarity matrix of other groups in the previous convolutional layer includes the intermediate temporal similarity matrix of group 1 in the first convolutional layer and the intermediate temporal similarity matrix of group 3 in the first convolutional layer.

[0082] Specifically, for all convolutional layers except the first one, when calculating the intermediate temporal similarity matrix of each convolutional layer, the input of each convolutional layer is the intermediate temporal similarity matrix of the previous convolutional layer, the frequency domain similarity matrix of the previous convolutional layer, and the intermediate temporal similarity matrix of the other groups of the previous convolutional layer. The output of the convolutional layer is the intermediate temporal similarity matrix of that convolutional layer.

[0083] S307. Obtain the intermediate temporal similarity matrix of the previous convolutional layer, the intermediate frequency similarity matrix of the previous convolutional layer, and the intermediate frequency similarity matrix of other groups of the previous convolutional layer through the other convolutional layers except the first convolutional layer, and determine the intermediate frequency similarity matrix of the other convolutional layers.

[0084] The intermediate frequency domain similarity matrix of other groups in the previous convolutional layer refers to the intermediate frequency domain similarity matrix output by the previous convolutional layer for other similarity matrix groups. For example, the similarity matrix groups include group 1, group 2, and group 3. During the calculation of group 2, when calculating the intermediate frequency domain similarity matrix of the second convolutional layer, the intermediate frequency domain similarity matrix of other groups in the previous convolutional layer includes the intermediate frequency domain similarity matrix of group 1 in the first convolutional layer and the intermediate frequency domain similarity matrix of group 3 in the first convolutional layer.

[0085] Specifically, for all convolutional layers except the first one, when calculating the intermediate frequency domain similarity matrix of each convolutional layer, the input of each convolutional layer is the intermediate time domain similarity matrix of the previous convolutional layer, the frequency domain similarity matrix of the previous convolutional layer, and the intermediate frequency domain similarity matrix of the other groups of the previous convolutional layer. The output of the convolutional layer is the intermediate frequency domain similarity matrix of that convolutional layer.

[0086] S308. The intermediate temporal similarity matrix of the last convolutional layer in the concatenated multiple convolutional layers is determined as the candidate temporal similarity matrix of the similarity matrix group, and the intermediate frequency domain similarity matrix of the last convolutional layer is determined as the candidate frequency domain similarity matrix of the similarity matrix group.

[0087] Specifically, for each similarity matrix group, the intermediate time-domain similarity matrix of the last convolutional layer in the similarity matrix group is determined as the candidate time-domain similarity matrix, and the intermediate frequency-domain similarity matrix of the last convolutional layer is determined as the candidate frequency-domain similarity matrix.

[0088] S309. Based on the weights, the candidate time-domain similarity matrix and the candidate frequency-domain similarity matrix of the similarity matrix group are weighted to determine the fusion result of the similarity matrix group.

[0089] S310. Based on the fusion results of each of the similarity matrix groups, determine at least one separation matrix, the number of which is the same as the number of pure signals included in the mixed signal.

[0090] S311. The encoding result and each of the separation matrices are fused together and decoded to obtain the separated pure signal.

[0091] The technical solution of this invention extracts features from the initial temporal similarity matrix and the initial frequency similarity matrix through multiple convolutional layers connected in series. In the similarity feature matrix group, the convolutional layers other than the first convolutional layer, when calculating the intermediate temporal similarity matrix, take the intermediate temporal similarity matrix of the previous convolutional layer, the intermediate frequency similarity matrix of the previous convolutional layer, and the intermediate temporal similarity matrices of other groups in the previous convolutional layer as input. This considers the influence of the intermediate frequency similarity matrix within the similarity matrix group and the intermediate temporal similarity matrix of other groups on the intermediate temporal similarity matrix, which can improve the accuracy of determining the candidate temporal similarity matrix. Similarly, the accuracy of determining the candidate frequency similarity matrix can also be improved, thereby improving the accuracy of the separation matrix and the separation effect of the mixed signal.

[0092] Example 4

[0093] Figure 4 This is a flowchart of a mixed signal separation method provided in Embodiment 4 of the present invention. Based on the above embodiments, this embodiment specifies the acquisition of the weights of the mixed signal as follows: acquiring the mixed signal; calculating the time-domain information density and frequency-domain information density of the mixed signal; inputting the time-domain information density and frequency-domain information density into a fully connected network to acquire the weights of the mixed signal. Figure 4 As shown, the method includes:

[0094] S410, Acquire the mixed signal.

[0095] Specifically, you can directly obtain the aliased mixed signal, or you can first alias at least two pure signals and then use the mixed signal as the mixed signal.

[0096] S420. Calculate the time-domain information density and frequency-domain information density of the mixed signal.

[0097] Time-domain information density refers to the information content of a mixed signal in the time domain. Frequency-domain information density refers to the information content of a mixed signal in the frequency domain. The lower the information content, the less information is available for signal separation, the more severe the aliasing, and the more difficult the signal separation. Conversely, the higher the information content, the more information is available for signal separation, the less severe the aliasing, and the easier the signal separation.

[0098] Specifically, the mixed signal undergoes a short-time Fourier transform (SFT) to obtain its time-frequency plot. The parameters of the SFT can be set according to the actual situation. The time-frequency plot describes the changes in the signal's frequency and amplitude over time. Based on the data in the time-frequency plot, the energy of the mixed signal in the time domain and frequency domain is calculated, forming a time-domain energy distribution plot and a frequency-domain energy distribution plot. Specifically, the data in the time-frequency plot is projected into the time and frequency domains respectively to obtain the time-domain energy distribution plot and the frequency-domain energy distribution plot. The data in the time-domain energy distribution plot and the frequency-domain energy distribution plot are normalized to obtain time-domain energy distribution data and frequency-domain energy distribution data. Information entropy is calculated for the time-domain energy distribution data and the frequency-domain energy distribution data respectively to obtain the time-domain information density and the frequency-domain information density. Information entropy is often used as a quantitative indicator of information content. It can be understood that the smaller the information entropy, the more concentrated the energy of the mixed signal, that is, the more severe the aliasing of the mixed signal, and the more difficult it is to separate the mixed signal. For example, the information entropy of the mixed signal in the time domain is less than that in the frequency domain, indicating that the aliasing of the mixed signal in the time domain is more severe than that in the frequency domain, and that separating the mixed signal in the time domain is more difficult than separating it in the frequency domain.

[0099] S430. Input the time-domain information density and frequency-domain information density into the fully connected network to obtain the weights of the mixed signal.

[0100] A fully connected network is essentially a single switch that connects all inputs and outputs. A fully connected layer refers to a layer where each node is connected to all nodes in the layer above it, used to synthesize extracted features. In this embodiment, a fully connected network may include at least one fully connected layer.

[0101] Specifically, the time-domain and frequency-domain information densities are input into the fully connected network, and the output of the fully connected network is used as the weights of the mixed signal. For example, the time-domain and frequency-domain information densities can be concatenated, and the concatenated result can be input into the fully connected network to obtain the weights of the mixed signal. These weights include both time-domain and frequency-domain weights.

[0102] S440: Obtain the encoding result of the mixed signal.

[0103] S450. Perform feature aggregation on the encoding result to obtain aggregated features, and determine at least one similarity matrix between the aggregated features and the encoding result based on the aggregated features and the encoding result.

[0104] S460. The similarity matrices are weighted by the weights to obtain at least one separation matrix.

[0105] S470. The encoding result and each of the separation matrices are fused together and decoded to obtain the separated pure signal.

[0106] The technical solution of this invention, by calculating the time-domain and frequency-domain information densities of the mixed signal, can determine the severity of aliasing in the time and frequency domains. Severe aliasing indicates difficulty in separating the mixed signal in that dimension. The fully connected network outputs smaller weights, allowing for dynamic adjustment of the weights when weighting the similarity matrix to obtain the separation matrix. This takes into account different aliasing situations in the time and frequency domains, improving the accuracy of the separation matrix.

[0107] Based on the above embodiments, the step of fusing the encoding result and each of the separation matrices and decoding them to obtain the separated pure signal includes: performing a dot product operation on the encoding result and each of the separation matrices to obtain at least one pure code; and decoding the pure code to obtain the separated pure signal.

[0108] Pure coding refers to the coding corresponding to the pure signal after separation. The number of pure codes is the same as the number of separation matrices. Specifically, the coding result can be regarded as a two-dimensional vector, and each separation matrix is ​​multiplied by it to obtain a pure code. A single pure code is obtained by multiplying the coding result by a separation matrix. These pure codes are then decoded by a decoder to obtain the separated pure signal.

[0109] Based on the above embodiments, after obtaining the separated pure signal, the method further includes: demodulating the pure signal to obtain the demodulated signal corresponding to the pure signal, thereby obtaining the information transmitted by the pure signal.

[0110] Demodulation refers to the process of recovering information from a modulated signal carrying information. In this application, a demodulator can be used to demodulate a clean signal. The demodulated signal is the signal after the clean signal has been demodulated.

[0111] Specifically, the separated pure signal is input into the demodulator to obtain the corresponding demodulated signal, and the information transmitted by the pure signal can be obtained from the demodulated signal.

[0112] By demodulating the pure signal, the mixed signal formed by the modulated signal can be separated to obtain the information transmitted in the mixed signal. This can expand the application range of the mixed signal separation method and reduce the dependence of the mixed signal separation method on the signal type.

[0113] Example 5

[0114] Figure 5aThis is a system structure diagram of a mixed signal separation method provided in Embodiment 5 of the present invention. The system of the mixed signal separation method provided in this embodiment can be trained using deep learning technology to improve the separation effect of the system, such as... Figure 5a As shown, the system includes an encoder, a splitter, and a decoder.

[0115] The encoder is used to encode the mixed signal to obtain the encoded result of the mixed signal. The encoder can be composed of multiple one-dimensional convolutional blocks.

[0116] The separator is used to determine the separation matrix of the mixed signal to achieve signal separation. The separator includes a waveform information extraction module, a feature aggregation module, and a separation matrix generation module. The waveform information extraction module obtains the weights of the mixed signal. The feature aggregation module performs feature aggregation on the encoding result to obtain aggregated features. The separator generation module generates at least one separation matrix based on the weights, aggregated features, and encoding result.

[0117] The decoder is used to decode the clean code to obtain the separated clean signal. Specifically, the decoding of the clean code can be achieved through multiple one-dimensional convolutions.

[0118] In this application, the SI-SNR (Scale-Invariant Source-To-Noise Ratio) of the purified signal obtained after the mixed signal is separated using a mixed signal separation method is used as the loss function. A pre-set training database is used as the input data for the system to train the entire system. During training, the parameters in the encoder, separator, and decoder are jointly optimized and updated using the loss function.

[0119] Specifically, for signals that are aliased in the time domain, C clean signals s can be estimated from the mixed signal in the time domain. i (t). Therefore, the mixed signal x(t) can be expressed as:

[0120] The mixed signal is input into the encoder to obtain the encoding result of the mixed signal, where the encoding result is w and B is the encoding basis function in the encoder. Then the mixed signal can be expressed as x = wB.

[0121] The encoded result w is actually a weighted sum of the pure codes of each pure signal, and its weight value is related to the proportion of each pure signal in the mixed signal. A separator can be used to obtain the separation matrix m of each pure signal in the encoded result w. i By performing a dot product operation with the mixed signal encoding result w, the pure code d of each pure signal can be obtained. i The calculation formula is: d i =mi ·w, where · represents the dot product of vectors. This yields the pure encoded vector d after separation of the source signals. i Then, the decoder processes the pure encoded d. i Decoding is performed by multiplying the signal with the decoder's basis function B to obtain the individual clean signal waveforms s. i This completes the signal separation task. The clean signal is represented as s. i =d i B.

[0122] Optionally, the system may also include a demodulator for demodulating the separated clean signal to obtain a demodulated signal.

[0123] In this embodiment of the invention, the separation matrix generation module includes at least one separation matrix calculator, which can ultimately output a fusion result of a similarity matrix group. The number of separation matrix calculators is an integer multiple of the number of clean signals after separation. Figure 5b A structural diagram of a classification matrix calculator is provided for Embodiment 5 of the present invention, as follows: Figure 5b As shown, similarity matrix calculations are performed in the time and frequency domains.

[0124] In a separation matrix calculator, a time-frequency permutation module transforms aggregated features and encoded results into the frequency domain. In the frequency domain, the aggregated features and encoded results are convolved twice, and then the convolution results of the aggregated features and encoded results are multiplied by a dot product to obtain an initial frequency domain similarity matrix. In the time domain, the aggregated features and encoded results are convolved twice, and then the convolution results of the aggregated features and encoded results are multiplied by a dot product to obtain a candidate initial domain similarity matrix.

[0125] The candidate temporal similarity matrix and candidate frequency similarity matrix are respectively input into multiple concatenated convolutional layers. The initial temporal similarity matrix is ​​convolved through the first convolutional layer to obtain the intermediate temporal similarity matrix of the first convolutional layer; the initial frequency similarity matrix is ​​convolved through the first convolutional layer to obtain the intermediate frequency similarity matrix of the first convolutional layer; through the other concatenated convolutional layers (excluding the first convolutional layer), the intermediate temporal similarity matrices of the previous convolutional layer, the intermediate frequency similarity matrix of the previous convolutional layer, and the intermediate temporal similarity matrices of other similarity matrix groups of the previous convolutional layer are obtained, thus determining the intermediate temporal similarity of the other convolutional layers. The similarity matrix is ​​obtained by taking the intermediate temporal similarity matrix of the previous convolutional layer, the intermediate frequency similarity matrix of the previous convolutional layer, and the intermediate frequency similarity matrix of the other similarity matrix groups of the previous convolutional layer through the other convolutional layers except the first convolutional layer, and determining the intermediate frequency similarity matrix of the other convolutional layers; the intermediate temporal similarity matrix of the last convolutional layer in the series of concatenated convolutional layers is determined as the candidate temporal similarity matrix of the similarity matrix group, and the intermediate frequency similarity matrix of the last convolutional layer is determined as the candidate frequency similarity matrix of the similarity matrix group.

[0126] Based on the weights of the mixed signals, the time-domain weights are applied to the candidate time-domain similarity matrices in the similarity matrix group, and the frequency-domain weights are applied to the candidate frequency-domain similarity matrices in the similarity matrix group. The weighted candidate time-domain similarity matrices and the weighted candidate frequency-domain similarity matrices are then added together to determine the fusion result of the similarity matrix group. Based on the fusion results of each similarity matrix group, at least one separation matrix is ​​determined, and the number of separation matrices is the same as the number of clean signals included in the mixed signal.

[0127] The technical solution of this invention obtains the weights and encoding results of the mixed signal, performs feature aggregation on the encoding results to obtain aggregated features, determines the similarity matrix between the aggregated features and the encoding results based on the aggregated features and the encoding results, and achieves the fusion of aggregated features and encoding results. At least one separation matrix is ​​obtained by weighting the similarity matrix with weights. After fusing the encoding results and the separation matrix, the fused result is decoded to obtain a clean signal. Different weights are obtained for different types of mixed signals. Considering the weights of the mixed signals during the process of obtaining the separation matrix enables the separation of different types of mixed signals, reducing the dependence on signal type when separating mixed signals, thereby improving the versatility of the mixed signal separation method. Simultaneously, the calculation of the separation matrix considers the influence of the encoding results, aggregated features, and weights on the calculation of the separation matrix, which can improve the accuracy of the separation matrix calculation and thus improve the separation effect of the mixed signal.

[0128] Example 6

[0129] Figure 6 This is a schematic diagram of a mixed signal separation device provided in Embodiment Six of the present invention. Figure 6 As shown, the device includes: an information acquisition module 601, a similarity matrix calculation module 602, a separation matrix calculation module 603, and a signal separation module 604.

[0130] Among them, the information acquisition module 601 is used to acquire the weights and encoding results of the mixed signal;

[0131] The similarity matrix calculation module 602 is used to perform feature aggregation on the encoding result to obtain aggregated features, and to determine at least one similarity matrix between the aggregated features and the encoding result based on the aggregated features and the encoding result;

[0132] The separation matrix calculation module 603 is used to weight each of the similarity matrices using the weights to obtain at least one separation matrix;

[0133] The signal separation module 604 is used to fuse the encoding result and each of the separation matrices, and then decode them to obtain the separated clean signal.

[0134] The technical solution of this invention obtains the weights and encoding results of the mixed signal, performs feature aggregation on the encoding results to obtain aggregated features, determines the similarity matrix between the aggregated features and the encoding results based on the aggregated features and the encoding results, and achieves the fusion of aggregated features and encoding results. At least one separation matrix is ​​obtained by weighting the similarity matrix with weights. After fusing the encoding results and the separation matrix, the fused result is decoded to obtain a clean signal. Different weights are obtained for different types of mixed signals. Considering the weights of the mixed signals during the process of obtaining the separation matrix enables the separation of different types of mixed signals, reducing the dependence on signal type when separating mixed signals, thereby improving the versatility of the mixed signal separation method. Simultaneously, the calculation of the separation matrix considers the influence of the encoding results, aggregated features, and weights on the calculation of the separation matrix, which can improve the accuracy of the separation matrix calculation and thus improve the separation effect of the mixed signal.

[0135] Optional, the similarity matrix calculation module 602 is specifically used for:

[0136] The similarity between the aggregated features and the encoding results is calculated in at least two dimensions to obtain at least two similarity matrices.

[0137] Optionally, the dimension includes the time domain and the frequency domain, and the similarity matrix includes: at least one initial time domain similarity matrix and at least one initial frequency domain similarity matrix, wherein the number of initial time domain similarity matrices is the same as the number of initial frequency domain similarity matrices;

[0138] Separation matrix calculation module 603 includes:

[0139] A grouping unit is used to group the similarity matrix to obtain at least one similarity matrix group, wherein a similarity matrix group includes an initial time-domain similarity matrix and an initial frequency-domain similarity matrix;

[0140] The candidate similarity matrix calculation unit is used to extract features from the initial time-domain similarity matrix and the initial frequency-domain similarity matrix of each similarity matrix group to obtain the candidate time-domain similarity matrix and the candidate frequency-domain similarity matrix of the similarity matrix group.

[0141] The fusion result determination unit is used to weight the candidate time-domain similarity matrix and the candidate frequency-domain similarity matrix of the similarity matrix group according to the weight, and determine the fusion result of the similarity matrix group;

[0142] A separation matrix determination unit is used to determine at least one separation matrix based on the fusion results of each of the similarity matrix groups, wherein the number of separation matrices is the same as the number of pure signals included in the mixed signal.

[0143] Optional, alternative similarity matrix calculation units include:

[0144] The first temporal intermediate matrix calculation subunit is used to convolve the initial temporal similarity matrix through the first convolutional layer among multiple concatenated convolutional layers to obtain the intermediate temporal similarity matrix of the first convolutional layer;

[0145] The first frequency domain intermediate matrix calculation subunit is used to convolve the initial frequency domain similarity matrix through the first convolutional layer to obtain the intermediate frequency domain similarity matrix of the first convolutional layer;

[0146] Other intermediate temporal matrix calculation subunits are used to obtain the intermediate temporal similarity matrix of the previous convolutional layer, the intermediate frequency similarity matrix of the previous convolutional layer, and the intermediate temporal similarity matrix of other groups of the previous convolutional layer through the other convolutional layers other than the first convolutional layer in the series of convolutional layers, and determine the intermediate temporal similarity matrix of the other convolutional layers.

[0147] Other frequency domain intermediate matrix calculation subunits are used to obtain the intermediate time domain similarity matrix of the previous convolutional layer, the intermediate frequency domain similarity matrix of the previous convolutional layer, and the intermediate frequency domain similarity matrix of other groups of the previous convolutional layer through other convolutional layers other than the first convolutional layer, and determine the intermediate frequency domain similarity matrix of the other convolutional layers.

[0148] The candidate similarity matrix determination subunit is used to determine the intermediate temporal similarity matrix of the last convolutional layer in the concatenated multiple convolutional layers as the candidate temporal similarity matrix of the similarity matrix group, and to determine the intermediate frequency domain similarity matrix of the last convolutional layer as the candidate frequency domain similarity matrix of the similarity matrix group.

[0149] Optionally, the information acquisition module 601 includes:

[0150] The signal acquisition unit is used to acquire mixed signals;

[0151] The information density calculation unit is used to calculate the time-domain information density and frequency-domain information density of the mixed signal;

[0152] The weight acquisition unit is used to input the time-domain information density and frequency-domain information density into the fully connected network to obtain the weights of the mixed signal.

[0153] Optionally, the signal separation module 604 includes:

[0154] A pure encoding determination unit is used to perform a dot multiplication operation between the encoding result and each separation matrix to obtain at least one pure encoding.

[0155] A clean signal acquisition unit is used to decode the clean code and acquire the separated clean signal.

[0156] Optionally, the device may also include:

[0157] The demodulation module is used to demodulate the pure signal, obtain the demodulated signal corresponding to the pure signal, and obtain the information transmitted by the pure signal.

[0158] The mixed signal separation device provided in the embodiments of the present invention can execute the mixed signal separation method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the method.

[0159] Example 7

[0160] Figure 7 A schematic diagram of an electronic device 10 that can be used to implement embodiments of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0161] like Figure 7 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 may also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0162] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0163] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as mixed-signal separation methods.

[0164] In some embodiments, the mixed-signal separation method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or mounted on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the mixed-signal separation method described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to perform the mixed-signal separation method by any other suitable means (e.g., by means of firmware).

[0165] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0166] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0167] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0168] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0169] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or computing systems that include middleware components (e.g., application servers), or computing systems that include frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0170] A computing system can include clients and servers. Clients and servers are generally geographically separated and typically interact via communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system. It addresses the shortcomings of traditional physical hosts and VPS (Virtual Private Server) services, such as high management difficulty and weak business scalability.

[0171] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0172] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A mixed-signal separation method, characterized in that, include: Obtain the weights and encoding results of the mixed signal; The encoding result is subjected to feature aggregation to obtain aggregated features, and at least one similarity matrix between the aggregated features and the encoding result is determined based on the aggregated features and the encoding result; By weighting each of the similarity matrices using the aforementioned weights, at least one separation matrix is ​​obtained; The encoding results and each of the separation matrices are fused together and then decoded to obtain the separated clean signal; Determining at least one similarity matrix between the aggregated features and the encoding results based on the aggregated features and the encoding results includes: The similarity between the aggregated features and the encoding results is calculated in at least two dimensions to obtain at least two similarity matrices; The dimension includes the time domain and the frequency domain, and the similarity matrix includes: at least one initial time domain similarity matrix and at least one initial frequency domain similarity matrix, wherein the number of initial time domain similarity matrices is the same as the number of initial frequency domain similarity matrices; The step of weighting each similarity matrix using the weights to obtain at least one separation matrix includes: The similarity matrix is ​​grouped to obtain at least one similarity matrix group, wherein a similarity matrix group includes an initial time-domain similarity matrix and an initial frequency-domain similarity matrix; For each similarity matrix group, feature extraction is performed on the initial time-domain similarity matrix and the initial frequency-domain similarity matrix of the similarity matrix group to obtain the candidate time-domain similarity matrix and the candidate frequency-domain similarity matrix of the similarity matrix group; Based on the weights, the candidate time-domain similarity matrix and the candidate frequency-domain similarity matrix of the similarity matrix group are weighted to determine the fusion result of the similarity matrix group; Based on the fusion results of each of the similarity matrix groups, at least one separation matrix is ​​determined, the number of which is the same as the number of pure signals included in the mixed signal.

2. The method according to claim 1, characterized in that, The step of extracting features from the initial time-domain similarity matrix and the initial frequency-domain similarity matrix of the similarity matrix group to obtain the candidate time-domain similarity matrix and the candidate frequency-domain similarity matrix of the similarity matrix group includes: The initial temporal similarity matrix is ​​convolved by the first convolutional layer among multiple convolutional layers to obtain the intermediate temporal similarity matrix of the first convolutional layer. The initial frequency domain similarity matrix is ​​convolved through the first convolutional layer to obtain the intermediate frequency domain similarity matrix of the first convolutional layer; By using the concatenated convolutional layers other than the first convolutional layer, the intermediate temporal similarity matrix of the previous convolutional layer, the intermediate frequency similarity matrix of the previous convolutional layer, and the intermediate temporal similarity matrix of other groups of the previous convolutional layer are obtained, and the intermediate temporal similarity matrix of the other convolutional layers is determined. By using the convolutional layers other than the first convolutional layer, the intermediate temporal similarity matrix of the previous convolutional layer, the intermediate frequency similarity matrix of the previous convolutional layer, and the intermediate frequency similarity matrix of other groups in the previous convolutional layer are obtained, and the intermediate frequency similarity matrix of the other convolutional layers is determined. The intermediate temporal similarity matrix of the last convolutional layer in the concatenated multiple convolutional layers is determined as the candidate temporal similarity matrix of the similarity matrix group, and the intermediate frequency domain similarity matrix of the last convolutional layer is determined as the candidate frequency domain similarity matrix of the similarity matrix group.

3. The method according to claim 1, characterized in that, The acquisition of the weights of the mixed signal includes: Acquire mixed signals; Calculate the time-domain and frequency-domain information densities of the mixed signal; The time-domain information density and frequency-domain information density are input into a fully connected network to obtain the weights of the mixed signal.

4. The method according to claim 1, characterized in that, The step of fusing the encoding result and each of the separation matrices, and then decoding them to obtain the separated clean signal includes: Perform a dot product operation between the encoding result and each separation matrix to obtain at least one pure encoding; The pure code is decoded to obtain the separated pure signal.

5. The method according to claim 1, characterized in that, After obtaining the separated pure signal, the process further includes: The clean signal is demodulated to obtain the demodulated signal corresponding to the clean signal, thereby obtaining the information transmitted by the clean signal.

6. A mixed signal separation device, characterized in that, include: The information acquisition module is used to acquire the weights and encoding results of the mixed signal; The similarity matrix calculation module is used to perform feature aggregation on the encoding result to obtain aggregated features, and determine at least one similarity matrix between the aggregated features and the encoding result based on the aggregated features and the encoding result; A separation matrix calculation module is used to weight each of the similarity matrices using the weights to obtain at least one separation matrix; The signal separation module is used to fuse the encoding result and each of the separation matrices, and then decode them to obtain the separated clean signal. The similarity matrix calculation module is specifically used for: The similarity between the aggregated features and the encoding results is calculated in at least two dimensions to obtain at least two similarity matrices; The dimension includes the time domain and the frequency domain, and the similarity matrix includes: at least one initial time domain similarity matrix and at least one initial frequency domain similarity matrix, wherein the number of initial time domain similarity matrices is the same as the number of initial frequency domain similarity matrices; The separation matrix calculation module includes: A grouping unit is used to group the similarity matrix to obtain at least one similarity matrix group, wherein a similarity matrix group includes an initial time-domain similarity matrix and an initial frequency-domain similarity matrix; The candidate similarity matrix calculation unit is used to extract features from the initial time-domain similarity matrix and the initial frequency-domain similarity matrix of each similarity matrix group to obtain the candidate time-domain similarity matrix and the candidate frequency-domain similarity matrix of the similarity matrix group. The fusion result determination unit is used to weight the candidate time-domain similarity matrix and the candidate frequency-domain similarity matrix of the similarity matrix group according to the weight, and determine the fusion result of the similarity matrix group; A separation matrix determination unit is used to determine at least one separation matrix based on the fusion results of each of the similarity matrix groups, wherein the number of separation matrices is the same as the number of pure signals included in the mixed signal.

7. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the mixed signal separation method according to any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute the mixed-signal separation method according to any one of claims 1-5.

Images