Long signal separation and fusion method, device and equipment of single-channel blind source separation model, medium and product

Abstract

The invention relates to the field of communication signal processing, and provides a long signal separation fusion method, device and equipment of a single-channel blind source separation model, a medium and a product, and the method comprises the steps: training the single-channel blind source separation model; making a to-be-separated aliasing signal slice sequence; performing reasoning separation operation on the to-be-separated aliasing signal slice sequence by using a trained single-channel blind source separation model to obtain two paths of separation component sequences and corresponding negative correlation flipping component sequences; performing front and back window component matching calculation of a sample separation frame on the two paths of separation component sequences and the corresponding negative correlation flipping component sequence to obtain an optimal matching component; based on the optimal matching component, performing window fusion calculation on the correct separation component of the multi-slice frame to obtain a fusion window component; and completing fusion reconstruction based on the fusion window component. According to the method, the problems of input limitation of a long signal on a single-channel blind source separation model and replaceability and overturning of a separation result can be solved at the same time.

Description

Technical Field

[0001] The present invention relates to the field of communication signal processing, and in particular to a long signal separation and fusion method, device, equipment, medium and product of a single-channel blind source separation model. Background Art

[0002] As the scope and means of utilizing electromagnetic resources continue to expand, spectrum crowding within the electromagnetic space and signal aliasing within the time-frequency space are becoming increasingly serious. Real-world communication signal processing often prioritizes addressing signal aliasing. For a single receiving station, the received signal is often a mixture of multiple signal components. The problem of separating such aliased signals from multiple sources at a single receiving station is often referred to as single-channel blind source separation (BSS), an extreme case of underdetermined BSS. With the recent development of deep learning methods, single-channel BSS algorithms based on single-channel BSS models such as Tasnet and Wave-U-Net have been initially applied and have achieved considerable progress.

[0003] After separating aliased signals, single-channel blind source separation models based on deep learning algorithms suffer from a permutation problem: the separated components output by the single-channel blind source separation model cannot be completely constrained to correspond one-to-one with the preset label sequence, and their positions may be swapped. Most researchers use permutation-invariant training (PIT) methods to ensure the correspondence between the loss and labels of the single-channel blind source separation model and the correctness of gradient updates during training. However, this still fails to constrain the label order of the components output by the single-channel blind source separation model. Because the loss function used in permutation-invariant training methods is typically based on the scale-invariant signal-to-noise ratio (SI-SNR) and its variants, the correlation between the separated components and the true component samples in the loss calculation process erases the sign difference between positive and negative correlations, resulting in the separation results of the single-channel blind source separation model sometimes being inverted (i.e., the separated signal waveform and the true standard sample waveform are flipped symmetrically about the zero axis). In real-world signal separation scenarios, the effective signal often reaches tens of thousands or even hundreds of thousands of points. Achieving end-to-end separation of such long sequences using deep learning single-channel blind source separation models is extremely difficult. Therefore, it is necessary to consider the conversion method from short sequence separation to long sequence separation. Summary of the Invention

[0004] Single-channel blind source separation of long signals requires solving both the input limitations of long signals on the single-channel blind source separation model and the permutation and flipping problems of the separation results. The present invention provides a long signal separation and fusion method, device, equipment, medium and product for a single-channel blind source separation model. By separating the original signal window slice frames, the single-channel blind source separation model reasoning of the long aliased signal is realized, and the accurate fusion of the front and back frames of the separation result is ensured to obtain an accurate long signal separation result.

[0005] In a first aspect, the present invention provides a long signal separation and fusion method for a single-channel blind source separation model, comprising:

[0006] Train a single-channel blind source separation model;

[0007] preparing a slice sequence of aliased signals to be separated;

[0008] Using the trained single-channel blind source separation model, an inference separation operation is performed on the aliased signal slice sequence to be separated to obtain two-way separation component sequences and a corresponding negative correlation flip component sequence;

[0009] Performing front and back window component matching calculations of the sample separation frame on the two-way separation component sequence and the corresponding negative correlation flip component sequence to obtain the optimal matching component;

[0010] Performing window fusion calculation on correctly separated components of the multi-slice frames based on the optimal matching components to obtain fused window components;

[0011] The fusion reconstruction is completed based on the fusion window components.

[0012] In some embodiments, the training of a single-channel blind source separation model includes:

[0013] The sample frame of the original aliased signal waveform is input as sample data into the single-channel blind source separation model. The output of the single-channel blind source separation model includes two separation components and corresponding negatively correlated flipped components.

[0014] During the training process, the SI-SNR loss is calculated, and the data pair combination used is generated by the four-way separation components output by the single-channel blind source separation model and the original two standard sample components input by the single-channel blind source separation model.

[0015] In some embodiments, the step of preparing a sequence of aliased signal slices to be separated comprises:

[0016] The time domain waveform data of long aliased signals of any length are sliced ​​according to the frame window length and step size to form a slicing sequence of aliased signals to be separated.

[0017] In some embodiments, performing matching calculation of front and back window components of the sample separation frame includes:

[0018] Taking the isolated component sequence A as an example, the i-th isolated component sepsA i , calculate it and sepsA respectively i+1 、sepsB i+1 sepsAinv i 、sepsBinv i+1 The waveform similarity index of the window overlapping part of the four separation components is calculated. According to the waveform similarity index of the window overlapping part, the separation component with the highest similarity is selected as the i-th separation component sepsA i The next moment window component of the window; wherein, the similarity calculation index is set to the mean square error of the waveform of the window overlapping part.

[0019] In some embodiments, performing window fusion calculation on correctly separated components of multi-slice frames based on the optimal matching components includes:

[0020] For multi-slice sequences, the optimal matching components of all overlapping windows are calculated according to the matching of the front and back window components of the sample separation frame, thereby obtaining the corresponding optimal matching component sequence;

[0021] The mean sequence formed by the intersection of the window component of the separated component at a certain moment and the waveform of the elements in the optimal matching component sequence in each time window with a width granularity of S is used as the fusion window component.

[0022] In some embodiments, performing fusion reconstruction based on the fusion window components includes:

[0023] For multi-slice sequences, after completing the optimal matching of all window components based on similarity using the window fusion algorithm of multi-frame separated components, the fusion of window components before and after the overlapping distance is performed in sequence according to the order of the windows, and finally the complete components of the two channels are reconstructed by the fusion of window slice components.

[0024] In a second aspect, the present invention provides a long signal separation and fusion device for a single-channel blind source separation model, comprising:

[0025] A first processing unit, used for training a single-channel blind source separation model;

[0026] A second processing unit, configured to prepare a slice sequence of aliased signals to be separated;

[0027] A third processing unit is configured to perform an inference separation operation on the aliased signal slice sequence to be separated using a trained single-channel blind source separation model to obtain a two-way separation component sequence and a corresponding negative correlation flipped component sequence;

[0028] a fourth processing unit, configured to perform front and rear window component matching calculation of the sample separation frame on the two separated component sequences and the corresponding negatively correlated flipped component sequence to obtain an optimal matching component;

[0029] A fifth processing unit is configured to perform window fusion calculation on the correctly separated components of the multi-slice frames based on the optimal matching components to obtain fused window components;

[0030] The sixth processing unit is configured to complete fusion reconstruction based on the fusion window components.

[0031] In a third aspect, the present invention provides an electronic device, comprising:

[0032] at least one processor; and a memory communicatively coupled to the at least one processor;

[0033] The memory stores instructions that can be executed by the at least one processor, and the at least one processor executes the above method by executing the instructions stored in the memory.

[0034] In a fourth aspect, the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium is used to store instructions, and when the instructions are executed, the above method is implemented.

[0035] In a fifth aspect, the present invention provides a computer program product, which, when called by a computer, enables the computer to execute the above method.

[0036] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0037] 1. It can improve the accuracy of sequence matching and fusion. The method of the present invention can automatically determine the window width and progressive step length. After adjustment, the wider the window overlap width, the higher the accuracy of the average in the algorithm. When the front and back window component matching algorithm of the sample separation frame is used to screen the correct separation component from the four separation components, when the overlap of the front window component is high, the correct one becomes in the positive sequence, otherwise it is negated, which can effectively improve the accuracy of sequence matching and fusion.

[0038] 2. It can solve the permutation and flipping problems caused by the training of a long signal single-channel blind source separation model. The present invention realizes the single-channel blind source separation model inference of a long aliased signal by separating the original signal window slice frames. It uses the front and back window component matching algorithm and the window fusion algorithm of the multi-frame separation components to ensure the accurate fusion of the front and back frames of the separation results, thereby achieving accurate long signal separation and solving the permutation and flipping problems caused by the training of a long signal single-channel blind source separation model.

[0039] 3. It can improve signal processing speed. The present invention optimizes the speed of window calculation, forward splicing, and reverse splicing used in processing long signals by selecting the correct ones before performing splicing. Compared with traditional methods based on code generation and reasoning, which have slow decoding speeds, speed tests show that the processing speed of the present method is increased by about 40 times. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 The present invention provides a flowchart of a long signal separation and fusion method for a single-channel blind source separation model.

[0041] Figure 2 A schematic structural diagram of a long signal separation and fusion device for a single-channel blind source separation model provided by an embodiment of the present invention.

[0042] Figure 3 A schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0043] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations.

[0044] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention as claimed, but rather merely represents selected embodiments of the present invention. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without creative effort are intended to fall within the scope of protection of the present invention.

[0045] Example

[0046] Single-channel blind source separation of long signals requires solving both the input restrictions of long signals on the single-channel blind source separation model and the permutation and flipping problems of the separation results. The embodiments of the present invention provide a long signal separation and fusion method, device, equipment, medium and product for a single-channel blind source separation model. By separating the original signal window slice frames, the single-channel blind source separation model reasoning of the long aliasing signal is realized, and the accurate fusion of the front and back frames of the separation result is ensured to obtain an accurate long signal separation result.

[0047] like Figure 1 As shown, an embodiment of the present invention provides a long signal separation and fusion method for a single-channel blind source separation model, comprising the following steps:

[0048] S100, training single-channel blind source separation model.

[0049] The sample frame of the original aliased signal waveform is input as sample data into the single-channel blind source separation model, and the frame window length is set to K (set as needed, for example, 256). The output of the single-channel blind source separation model includes two separation components SepA and SepB, and the corresponding negative correlation flip component SepA. * 、SepB * ;

[0050] During the training process, the SI-SNR loss is calculated, and the data pair combination used is the four-way separation components output by the single-channel blind source separation model (two-way separation components SepA, SepB and the corresponding negative correlation flip component SepA * 、SepB * ) is generated in pairs with the original two standard sample components LabelA and LabelB input by the single-channel blind source separation model to ensure that the loss value corresponding to each gradient update fully considers the impact of component flipping under negative correlation.

[0051] S200: Generate a slice sequence of aliased signals to be separated.

[0052] For the time domain waveform data of long aliased signals of arbitrary length L, slice them according to the frame window length K and step size S to form a aliased signal slice sequence to be separated, which is expressed as:

[0053] SignalSlice=[ss0,ss1,...,ss n ];

[0054] Among them, ss i (i=0,1,...,n) is a time series of length K,

[0055] S300, using a trained single-channel blind source separation model to perform an inference separation operation on the aliased signal slice sequence to be separated, to obtain two-way separation component sequences and a corresponding negative correlation flipped component sequence;

[0056] The two separation component sequences are respectively expressed as:

[0057] SepSliceA=[sepsA0,sepsA1,...,sepsA n ];

[0058] SepSliceB=[sepsB0,sepsB1,...,sepsB n ];

[0059] The corresponding negative correlation flip component sequences are respectively expressed as:

[0060] SepSliceAinverse=[sepsAinv0,sepsAinv1,...,sepsAinv n ];

[0061] SepSliceBinverse=[sepsBinv0,sepsBinv1,...,sepsBinv n ].

[0062] S400 , performing front and back window component matching calculation of the sample separation frame on the two separated component sequences and the corresponding negatively correlated flipped component sequence to obtain the optimal matching component.

[0063] For the four separated components formed after separation by the single-channel blind source separation model, it is necessary to consider the impact of the component flipping problem under negative correlation, and use the front and back window component matching algorithm of the sample separation frame to screen out the correct separated components from the four separated components. The front and back window component matching algorithm of the sample separation frame includes:

[0064] Considering the permutation and flipping of separation components, taking the separation component sequence A as an example, for the i-th separation component sepsA in SepSliceA i , calculate it and sepsA respectively i+1 、sepsB i+1 sepsAinv i 、sepsBinv i+1 The waveform similarity index of the window overlapping part of the four separation components is calculated. According to the waveform similarity index of the window overlapping part, the separation component with the highest similarity is selected as the i-th separation component sepsA i The similarity calculation index is set as the mean square error (RMSE) of the waveforms of the overlapping parts of the windows to ensure the absolute similarity of the waveforms.

[0065] Correspondingly, for the i-th separation component sepsB in SepSliceB i The next moment window component of is also determined in the above manner, but since the component has replacement exclusivity, that is, after the previous window component determines its next moment window component, the next moment window component of another window can only be the remaining next moment window component or the corresponding flipped component.

[0066] The matching algorithm of the front and back window components of the sample separation frame is recorded as BestMatch, then the i-th separation component sepsA of the separation component sequence A is i The optimal matching component is expressed as:

[0067] BMsepsliceA t= BestMatch(sepsA t , {sepsA t+1 , sepsB t+1 , sepsAinv t+1 , sepsBinv t+1});

[0068] S500, perform window fusion calculation on the correctly separated components of the multi-slice frame based on the optimal matching component to obtain the fused window component.

[0069] For the multi-slice sequence, apply the window fusion algorithm for multi-frame separated components to fuse and splice the selected correctly separated components to achieve the fused reconstruction of the long signal. The specific calculation method is as follows:

[0070] Assume K = q * S, where q is an integer not less than 1. For the window component sepslice of the separated component A at a given moment t , the subsequent set of window components that move with a step size and overlap with it is:

[0071] SC = {SS t+1 , SS t+2 ,..., SS t+q-1};

[0072] Among them, SS t = {sepsA t , sepsB t , sepsAinv t , sepsBinv t}; The overlapping lengths of each overlapping window component are (q - 1)s, (q - 2)s,..., s in sequence.

[0073] The first step is to calculate the optimal matching components of all overlapping windows:

[0074] For 1 ≤ i < q, where i is an integer, calculate the optimal matching component of the i-th overlapping window component SS t+i and the window component sepslice of the separated component A at a given moment t according to the front and rear window component matching algorithm of the sample separation frame:

[0075] BMsepslice t+i = BestMatch(sepsA t , SS t+i );

[0076] The overlapping length involved in the calculation is (q - i)s.

[0077] By the window component sepslice of the separated component A at a given moment tCalculate the optimal matching component for all elements in the overlapping window component set SC and obtain the corresponding optimal matching component sequence, which is expressed as:

[0078] BMall t ={BMsepslice t+1 ,BMsepslice t+2 ,...,BMsepslice t+q-1}.

[0079] The second step is to calculate the window component sepslice of the separation component A at a certain moment t and the optimal matching component sequence BMall t The mean sequence of the waveform intersection of the internal elements in each time window with a width granularity of S is used as the fusion window component. As a method of repeated iterative calculation, the window fusion method only calculates the window component sepslice of the separation component A at a certain moment t The width granularity S after the end point, the waveform outside this range is not considered and is left for estimation in the next iterative calculation. Within the calculation range, the window component sepslice of the separated component A at a certain moment t The weighted average is performed on the best matching component according to the width granularity S to obtain the fusion window component.

[0080] S600: Complete fusion reconstruction based on the fusion window components.

[0081] Specifically, for multi-slice sequences, after completing the similarity-based optimal matching of all window components using the window fusion algorithm of multi-frame separated components, the fusion of window components before and after the overlapping distance is performed in sequence according to the order of the windows, and finally the complete components of the two channels are reconstructed by the fusion of window slice components.

[0082] Based on the same technical concept, such as Figure 3 As shown, an embodiment of the present invention provides a long signal separation and fusion device for a single-channel blind source separation model, comprising:

[0083] The first processing unit is used to calibrate the observation system error and thereby design an error tolerance value;

[0084] The second processing unit is used to collect the observation time difference of the observation system;

[0085] The third processing unit is used to select an observation range and establish a discretized map;

[0086] a fourth processing unit, configured to traverse each cell in the discretized map and assign a value based on the error tolerance value and the observation time difference;

[0087] The fifth processing unit is used to traverse each cell on the map again after completing the assignment of values ​​to all cells, and find the cell with the largest value as the positioning result.

[0088] As for the specific processing methods of each processing unit in the above-mentioned device, reference can be made to the specific description of the above-mentioned method, which will not be repeated here.

[0089] Based on the same technical concept, an embodiment of the present invention further provides an electronic device that can implement the long signal separation and fusion method process of the single-channel blind source separation model provided in the above embodiment of the present invention. In one embodiment, the electronic device can be a server, or a terminal device or other electronic device. Figure 3 As shown, the electronic device may include:

[0090] At least one processor, and a memory connected to the at least one processor. The embodiment of the present invention does not limit the specific connection medium between the processor and the memory. Figure 3 The example in this article is that the processor and memory are connected via a bus. Figure 3 The connections between the other components are shown in bold lines, which are only for illustration and not intended to be limiting. The bus can be divided into address bus, data bus, control bus, etc. Figure 3 The processor is represented by a single thick line, but this does not mean that there is only one bus or only one type of bus. Alternatively, the processor can also be called a controller, without any limitation on the name.

[0091] In an embodiment of the present invention, the memory stores instructions that can be executed by at least one processor. The at least one processor can execute the long signal separation and fusion method of the single-channel blind source separation model discussed above by executing the instructions stored in the memory. The processor can implement Figure 3 The functions of each module in the device shown.

[0092] Among them, the processor is the control center of the device, which can use various interfaces and lines to connect the various parts of the entire control device, and monitor the device as a whole by running or executing instructions stored in the memory and calling data stored in the memory, the various functions of the device and processing data.

[0093] In an optional design, the processor may include one or more processing units, and the processor may integrate an application processor and a modem processor, wherein the application processor primarily processes the operating system, user interface, and application programs, and the modem processor primarily processes wireless communications. It is understood that the modem processor may not be integrated into the processor. In some embodiments, the processor and memory may be implemented on the same chip, or in some embodiments, they may be implemented on separate chips.

[0094] The processor can be a general-purpose processor, such as a CPU, a digital signal processor, an application-specific integrated circuit, a field-programmable gate array or other programmable logic device, a discrete gate or transistor logic device, or a discrete hardware component, and can implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of the present invention. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the long signal separation and fusion method for a single-channel blind source separation model disclosed in the embodiments of the present invention can be directly implemented as a hardware processor, or can be implemented using a combination of hardware and software modules in the processor.

[0095] As a non-volatile computer-readable storage medium, memory can be used to store non-volatile software programs, non-volatile computer executable programs and modules. Memory can include at least one type of storage medium, for example, can include flash memory, hard disk, multimedia card, card-type memory, random access memory (Random Access Memory, RAM), static random access memory (Static Random Access Memory, SRAM), programmable read-only memory (Programmable Read Only Memory, PROM), read-only memory (Read Only Memory, ROM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, EEPROM), magnetic memory, disk, optical disk, etc. Memory is any other medium that can be used to carry or store desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory in the embodiment of the present invention can also be a circuit or any other device that can realize a storage function, for storing program instructions and / or data.

[0096] By designing and programming the processor, the code corresponding to the long signal separation and fusion method for a single-channel blind source separation model described in the aforementioned embodiment can be embedded in the chip, thereby enabling the chip to execute the steps of the method of the aforementioned embodiment during operation. Designing and programming the processor is well known to those skilled in the art and will not be further described here.

[0097] Based on the same inventive concept, an embodiment of the present invention further provides a storage medium storing computer instructions. When the computer instructions are executed on a computer, the computer executes the long signal separation and fusion method of a single-channel blind source separation model discussed above.

[0098] In some optional embodiments, the present invention also provides various aspects of a long signal separation and fusion method for a single-channel blind source separation model, which can also be implemented in the form of a program product, which includes program code. When the program product is run on an apparatus, the program code is used to enable the control device to execute the steps of a long signal separation and fusion method for a single-channel blind source separation model according to various exemplary embodiments of the present invention described above in this specification.

[0099] It should be noted that although several units or subunits of the device are mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to an embodiment of the present invention, the features and functions of two or more units described above can be embodied in one unit. Conversely, the features and functions of a unit described above can be further divided into multiple units to be embodied. In addition, although the operations of the method of the present invention are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in this specific order, or that all the operations shown must be performed to achieve the desired results. Additionally or alternatively, certain steps can be omitted, multiple steps can be combined into one step, and / or one step can be decomposed into multiple steps.

[0100] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0101] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as a combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a server, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the process in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0102] Program code for performing the operations of the present invention may be written using any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computing device, partially on the user's device, as a stand-alone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0103] Where a remote computing device is involved, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., through the Internet using an Internet service provider).

[0104] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0105] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0106] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A long signal separation and fusion method for a single-channel blind source separation model, characterized in that: include: Train a single-channel blind source separation model; preparing a slice sequence of aliased signals to be separated; Using the trained single-channel blind source separation model, an inference separation operation is performed on the aliased signal slice sequence to be separated to obtain two-way separation component sequences and a corresponding negative correlation flip component sequence; Performing front and back window component matching calculations on the two separated component sequences and the corresponding negatively correlated flipped component sequences of the sample separation frame to obtain the optimal matching component; Performing window fusion calculation on correctly separated components of the multi-slice frames based on the optimal matching components to obtain fused window components; The fusion reconstruction is completed based on the fusion window components.

2. The long signal separation and fusion method of the single-channel blind source separation model according to claim 1 is characterized in that: The training single-channel blind source separation model includes: The sample frame of the original aliased signal waveform is input as sample data into the single-channel blind source separation model. The output of the single-channel blind source separation model includes two separation components and corresponding negatively correlated flipped components. During the training process, the SI-SNR loss is calculated, and the data pair combination used is generated by the four-way separation components output by the single-channel blind source separation model and the original two standard sample components input by the single-channel blind source separation model.

3. The long signal separation and fusion method of the single-channel blind source separation model according to claim 1 is characterized in that: The step of preparing a slice sequence of aliased signals to be separated comprises: The time domain waveform data of long aliased signals of any length are sliced ​​according to the frame window length and step size to form a slicing sequence of aliased signals to be separated.

4. The long signal separation and fusion method of the single-channel blind source separation model according to claim 1 is characterized in that: The front and rear window component matching calculation of the sample separation frame includes: Taking the isolated component sequence A as an example, the i-th isolated component sepsA i , calculate it and sepsA respectively i+1 、sepsB i+1 、sepsAinv i 、sepsBinv i+1 The waveform similarity index of the window overlapping part of the four separation components is calculated. According to the waveform similarity index of the window overlapping part, the separation component with the highest similarity is selected as the i-th separation component sepsA i The next moment window component of the window; wherein, the similarity calculation index is set to the mean square error of the waveform of the window overlapping part.

5. The long signal separation and fusion method of the single-channel blind source separation model according to claim 4 is characterized in that, The performing window fusion calculation on the correctly separated components of the multi-slice frames based on the optimal matching components includes: For multi-slice sequences, the optimal matching components of all overlapping windows are calculated according to the matching of the front and back window components of the sample separation frame, thereby obtaining the corresponding optimal matching component sequence; The mean sequence formed by the intersection of the window component of the separated component at a certain moment and the waveform of the elements in the optimal matching component sequence in each time window with a width granularity of S is used as the fusion window component.

6. The long signal separation and fusion method of the single-channel blind source separation model according to claim 5 is characterized in that: The completing fusion reconstruction based on the fusion window components includes: For multi-slice sequences, after completing the optimal matching of all window components based on similarity using the window fusion algorithm of multi-frame separated components, the fusion of window components before and after the overlapping distance is performed in sequence according to the order of the windows, and finally the complete components of the two channels are reconstructed by the fusion of window slice components.

7. A long signal separation and fusion device for a single-channel blind source separation model, characterized in that: include: A first processing unit, used for training a single-channel blind source separation model; A second processing unit, configured to prepare a slice sequence of aliased signals to be separated; A third processing unit is configured to perform an inference separation operation on the aliased signal slice sequence to be separated using a trained single-channel blind source separation model to obtain a two-way separation component sequence and a corresponding negative correlation flipped component sequence; a fourth processing unit, configured to perform front and rear window component matching calculation of the sample separation frame on the two separated component sequences and the corresponding negatively correlated flipped component sequence to obtain an optimal matching component; A fifth processing unit is configured to perform window fusion calculation on the correctly separated components of the multi-slice frames based on the optimal matching components to obtain fused window components; The sixth processing unit is configured to complete fusion reconstruction based on the fusion window components.

8. An electronic device, characterized in that: include: at least one processor; and a memory communicatively coupled to the at least one processor; The memory stores instructions that can be executed by the at least one processor, and the at least one processor executes the method according to any one of claims 1 to 6 by executing the instructions stored in the memory.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium is used to store instructions, and when the instructions are executed, the method according to any one of claims 1 to 6 is implemented.

10. A computer program product, characterized in that When the computer program product is called by a computer, the computer is caused to execute the method according to any one of claims 1 to 6.

Images