Blind source separation method based on self-encoder combined rapid independent component analysis

Abstract

The invention provides a blind source separation method based on self-encoder joint fast independent component analysis. The method comprises the following steps: constructing and training a # imgabs0 # network based on a self-encoder; carrying out blind source separation on the to-be-separated aliasing signal to obtain a recovered original independent signal; the # imgabs1 # network comprises a # imgabs2 # network, a # imgabs3 # network and a # imgabs4 # network; the training comprises the following steps: generating an aliasing signal, and obtaining a to-be-recovered original independent signal after blind source separation of the aliasing signal; inputting the aliasing signal into a # imgabs5 network to obtain a potential space representation, fusing the potential space representation with the to-be-recovered original independent signal, and inputting the fused signal into a # imgabs6 network to obtain a recovered original independent signal; and calculating the loss of the recovered original independent signal and the original independent signal to be aliased, carrying out back propagation, and iteratively optimizing the parameters of the # imgabs7 # network. According to the invention, the separation accuracy of aliasing signals is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of blind source separation, and in particular to a blind source separation method based on an autoencoder combined with fast independent component analysis. Background Art

[0002] Currently, people usually use a blind source separation (BSS) algorithm to separate unobserved original signals from multiple observed mixed signals.

[0003] Among them, the blind source separation algorithm is often used to solve the famous "cocktail party problem" in the field of computer speech recognition. This problem describes a cocktail party where multiple people speak at the same time, and multiple recording devices record at the same time, and then use relevant algorithms to identify the speech content after the meeting. However, current speech recognition technology can identify the speech content of a single person with high accuracy. For scenarios where multiple people speak at the same time, the accuracy of speech recognition will drop significantly. For scenarios where multiple people speak at the same time, it is necessary to restore the independent original speech of a single person based on the voice-aliased recording file in order to more accurately identify the independent original speech content.

[0004] Secondly, with the development of communication technology, a large number of military and civilian electronic equipment have been put into use, resulting in a diversity of signal modulation styles, which in turn constitutes a complex electromagnetic environment. Especially for non-cooperative broadband reception, the observed signals are often unknown and aliased in the time and frequency domains. Therefore, blind source separation algorithms are also often used to recover the original independent signals based on the aliased signals received by the communication system.

[0005] Independent Component Analysis (ICA) algorithm is the mainstream method for blind source signal separation. In 1996, Finnish scholars first proposed the Fast Independent Component Analysis (Fast-ICA) algorithm based on degree, and in 1999 proposed the improved Fast-ICA algorithm based on negative entropy. This algorithm decomposes multi-channel observation signals into several independent components according to the principle of statistical independence and through an optimization algorithm. Although the Fast-ICA algorithm can decompose the aliased signal into multiple unrelated original independent signals, it has two uncertainties: the poor accuracy of the amplitude of the separated original independent signal and the phase reversal, that is, the accuracy of the separated original independent signal is poor.

[0006] In order to solve the above problems, people have been seeking an ideal technical solution. Summary of the invention

[0007] Based on this, it is necessary to provide a blind source separation method based on autoencoder combined with fast independent component analysis to solve the above technical problems. The present invention repairs the amplitude and phase inversion defects of the original independent signal obtained by the Fast-ICA algorithm by constructing an AE-ICA network based on an autoencoder (AE), so that the repaired signal has higher accuracy.

[0008] In order to achieve the above object, the present invention provides a blind source separation method based on autoencoder combined with fast independent component analysis, which comprises:

[0009] Build and train an autoencoder-based AE-ICA network;

[0010] The AE-ICA network is used to perform blind source separation on the input aliased signal to be separated, and the restored original independent signal is obtained;

[0011] The AE-ICA network based on the autoencoder includes an Encoder network, a Fast-ICA network and a Decoder network;

[0012] The method of training the autoencoder-based AE-ICA network includes:

[0013] (1) generating an aliased signal based on an original independent signal to be aliased;

[0014] (2) inputting the aliased signal into the Fast-ICA network for blind source separation to obtain the original independent signal to be restored;

[0015] (3) inputting the aliased signal into the Encoder network to obtain a latent space representation of the aliased signal;

[0016] (4) fusing the latent space representation with the original independent signal to be restored, obtaining a fused signal and inputting it into the Decoder network to obtain the restored original independent signal;

[0017] (5) calculating the loss of the restored original independent signal and the original independent signal to be aliased and performing back propagation to optimize the parameters of the autoencoder-based AE-ICA network;

[0018] (6) Iteratively optimize the parameters of the autoencoder-based AE-ICA network until the number of iterations reaches a preset number and then stops iterating, or, based on the difference between the restored original independent signal corresponding to the current iteration and the original independent signal to be aliased, determine whether to continue iteratively optimizing the parameters of the autoencoder-based AE-ICA network.

[0019] The beneficial effects of the present invention are:

[0020] The present invention fuses the latent space representation and the corresponding original independent signal to be restored to obtain a fused signal, and then inputs the fused signal into a decoder (Decoder network) of an autoencoder to obtain the restored original independent signal, and then iteratively optimizes the parameters of the autoencoder according to the loss back-propagation of the restored original independent signal and the corresponding original independent signal to be aliased, so as to obtain an AE-ICA network based on the autoencoder with higher accuracy in separating aliased signals; finally, according to the trained AE-ICA network based on the autoencoder, the original independent signal of the aliased signal to be separated can be obtained more accurately, the amplitude of the separated original independent signal is more accurate, and the phase of the separated original independent signal is more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a simplified schematic diagram of the AE-ICA network structure based on the autoencoder of the present invention;

[0022] Figure 2 It is a simplified schematic diagram of the Encoder network structure of the autoencoder of the present invention;

[0023] Figure 3 It is a simplified schematic diagram of the Decoder network structure of the autoencoder of the present invention;

[0024] Figure 4 It is a schematic diagram of two original modulation signals of the present invention;

[0025] Figure 5 It is a schematic diagram of the result of the first aliasing of two original modulation signals of the present invention;

[0026] Figure 6 is a schematic diagram of the result of the second aliasing of two original modulation signals of the present invention;

[0027] Figure 7 It is a schematic diagram for comparing the original modulation signal 1 of the present invention, the original independent signal corresponding to the original modulation signal 1 separated by the Fast-ICA algorithm, and the original independent signal corresponding to the original modulation signal 1 separated by the AE-ICA algorithm of the present invention;

[0028] Figure 8 It is a schematic diagram for comparing the original modulated signal 2 of the present invention, the original independent signal corresponding to the original modulated signal 2 separated by the Fast-ICA algorithm, and the original independent signal corresponding to the original modulated signal 2 separated by the AE-ICA algorithm of the present invention;

[0029] Fig. 9It is a schematic diagram comparing the original modulation signal 3 of the present invention, the original independent signal corresponding to the original modulation signal 3 separated by the Fast-ICA algorithm, and the original independent signal corresponding to the original modulation signal 3 separated by the AE-ICA algorithm of the present invention. DETAILED DESCRIPTION

[0030] The technical solution of the present invention is further described in detail below through specific implementation methods.

[0031] The terms "first", "second", "third", "fourth", etc. in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the terms used in this way can be interchangeable under appropriate circumstances. This is just a way of distinguishing objects with the same attributes when describing the embodiments of this application.

[0032] For ease of understanding, the interactive parties and / or terms and / or custom words involved in the present invention are first described in conjunction with the technical solution of the present invention:

[0033] Autoencoder (AE) is a deep learning model that can learn the feature representation of data through unsupervised learning. The autoencoder consists of an encoder and a decoder, where the encoder maps the input data to a low-dimensional latent space and the decoder reconstructs the representation of the latent space into the original input data.

[0034] It should be noted that the present invention is based on the AutoEncoder (AE) widely used in deep learning to compensate for the defects of the Fast-ICA algorithm. After the defect compensation, the amplitude accuracy of the separated original independent signals is higher and the phase accuracy is higher.

[0035] This embodiment provides a specific implementation method of a blind source separation method based on an autoencoder combined with fast independent component analysis. The network structure designed by the present invention is as follows: Figure 1 As shown in Figure 1, the main body of the network consists of two parts: the autoencoder and the Fast-ICA, where x(t) represents the aliased signal. represents the separated original independent signal. The method comprises:

[0036] Build and train an autoencoder-based AE-ICA network;

[0037] The AE-ICA network is used to perform blind source separation on the input aliased signal to be separated, and the restored original independent signal is obtained;

[0038] The AE-ICA network based on the autoencoder includes an Encoder network, a Fast-ICA network and a Decoder network;

[0039] The method of training the autoencoder-based AE-ICA network includes:

[0040] (1) generating an aliased signal based on an original independent signal to be aliased;

[0041] (2) inputting the aliased signal into the Fast-ICA network for blind source separation to obtain the original independent signal to be restored;

[0042] (3) inputting the aliased signal into the Encoder network to obtain a latent space representation of the aliased signal;

[0043] (4) fusing the latent space representation with the original independent signal to be restored, obtaining a fused signal and inputting it into the Decoder network to obtain the restored original independent signal;

[0044] (5) calculating the loss of the restored original independent signal and the original independent signal to be aliased and performing back propagation to optimize the parameters of the autoencoder-based AE-ICA network;

[0045] (6) Iteratively optimize the parameters of the autoencoder-based AE-ICA network until the number of iterations reaches a preset number and then stops iterating, or, based on the difference between the restored original independent signal corresponding to the current iteration and the original independent signal to be aliased, determine whether to continue iteratively optimizing the parameters of the autoencoder-based AE-ICA network.

[0046] It should be noted that the present invention first optimizes the parameters of the autoencoder, and then uses the trained AE-ICA network based on the autoencoder to separate the aliased signal to be separated, so as to obtain the corresponding restored original independent signal, so as to improve the accuracy of the amplitude and phase of the separated original independent signal.

[0047] In some embodiments, (1) generating an aliased signal based on an original independent signal to be aliased includes: generating the aliased signal x(t) using the following formula:

[0048]

[0049] Where n is the number of signal sources (i.e. the total number of original independent signals to be aliased), m is the number of receiving channels (i.e. the total number of aliased signals received), v(t) is additive white Gaussian noise (AWGN), and T is the signal length. In order to make the above formula simpler, the above formula is abbreviated as x(t)=As(t)+v(t). Where A∈R m×n , A is the linear aliasing matrix, s1(t), s2(t)…s n (t) is n independent modulation signals (in other words, n is the total number of independent modulation signal sources).

[0050] In some embodiments, n=2, m=2.

[0051] In some embodiments, before the aliasing signal is input into the Fast-ICA network, it also includes: decentralizing the aliasing signal so that the mean of the processed signal is 0; and performing "whitening" preprocessing on the aliasing signal. The purpose of decentralization is to simplify and facilitate subsequent calculations. The purpose of whitening preprocessing is to remove the correlation between the aliasing signals, simplify the subsequent extraction process of the original independent signal, and improve the convergence of the algorithm.

[0052] It should be noted that the aliased signal input to the encoder (Encoder network) of the autoencoder is not preprocessed (ie, no decentralization and whitening preprocessing is performed).

[0053] In some embodiments, the Fast-ICA algorithm comes from Scikit-learn, a popular modular machine learning library in the Python environment, and specifically from the sklearn.decomposition.FastICA module. It should be noted that the module already includes whitening preprocessing.

[0054] (2) inputting the aliased signal into the Fast-ICA network for blind source separation to obtain the original independent signal to be restored, including:

[0055] S ICA =FastICA[x(t)],{t=1,2,…,T}

[0056] Among them, S ICA represents the original independent signal to be restored (in the form of a matrix), and FastICA[] represents the Fast-ICA algorithm (Fast-ICA network).

[0057] (3) Inputting the aliased signal into the Encoder network to obtain a latent space representation of the aliased signal, including:

[0058] F enc =Enc[x(t)],{t=1,2,…,T}

[0059] F enc Represents the latent space representation of the aliased signal (in the form of a matrix), and Enc[] represents the encoder (Encoder network).

[0060] like Figure 2 and Figure 3 As shown, in some embodiments, the Encoder network includes a first LSTM network (Long Short-Term Memory Network), a linear layer Linear, an activation function ReLU and a linear layer Linear connected in sequence; the first LSTM network is LSTM1;

[0061] The first LSTM network includes 3 layers of LSTM, the input dimension is the number of receiving channels 2, and the hidden layer is 256;

[0062] The Decoder network includes a linear layer Linear, an activation function ReLU, a linear layer Linear, a second LSTM network (LSTM2) and a linear layer Linear connected in sequence;

[0063] The second LSTM network includes 3 layers of LSTM, the input dimension is 2, the number of receiving channels, and the hidden layer is 256.

[0064] In some embodiments, in order to fix the defects of the Fast-ICA algorithm, (4) the latent space representation is fused with the original independent signal to be restored to obtain a fused signal. The fusion of the latent space representation and the original independent signal to be restored is a dot product fusion. For example, the calculation formula of the fused signal is:

[0065] F=F enc ⊙S ICA

[0066] Among them, F represents the fusion signal, ⊙ represents the matrix dot product, and F enc represents the latent space representation, S ICA Represents the original independent signal to be restored.

[0067] The fused signal is input into the Decoder network to obtain the restored original independent signal, including: S ′ =Dec[F] =Dec[F enc ⊙S ICA ], where S ′ represents the restored original independent signal (i.e., the original independent signal split by the decoder), S ′is in the form of a matrix, Dec[] represents the decoder (Decoder network), and F represents the fused signal (i.e., the fused latent space representation). This step uses the decoder to remap the fused signal to the original independent signal.

[0068] In some implementations, the loss of the restored original independent signal and the original independent signal to be aliased is The calculation formula is:

[0069]

[0070] Wherein, s(t) is denoted as S, S represents the original independent signal to be aliased, ∥∥2 represents the vector 2-norm (L2 norm), n represents the total number of original independent signals to be aliased, and i represents the index of the original independent signal.

[0071] It represents the mean of the squares of the vector 2-norm values ​​of each restored original independent signal and the corresponding original independent signal to be aliased.

[0072] It should be noted that the smaller the loss (i.e., difference) between the restored original independent signal and the original independent signal to be aliased, the greater the accuracy of the parameters of the AE-ICA network based on the autoencoder. Then, back propagation is selected based on the loss (i.e., difference) between the restored original independent signal and the original independent signal to be aliased to optimize the parameters of the AE-ICA network based on the autoencoder.

[0073] It should be noted that the technical means of performing back propagation according to the loss function to optimize the parameters of the autoencoder is an existing technology and will not be described in detail.

[0074] In some embodiments, the parameters of the autoencoder-based AE-ICA network are iteratively optimized until the number of iterations reaches a preset number (which can be 50 times) and the iteration is stopped. Those skilled in the art can set the preset number of times. Through multiple iterations, a trained autoencoder-based AE-ICA network can be obtained, and then the defects of the Fast-ICA algorithm can be repaired according to the trained autoencoder-based AE-ICA network.

[0075] In some embodiments, whether to continue iteratively optimizing the parameters of the autoencoder-based AE-ICA network is determined based on the difference between the restored original independent signal corresponding to the current iteration (ie, the latest one) and the original independent signal to be aliased.

[0076] Specifically, in some embodiments, in response to the difference between the original independent signal recovered corresponding to the current iteration and the original independent signal to be aliased being less than a preset threshold, the iterative optimization of the parameters of the AE-ICA network based on the autoencoder is stopped. Those skilled in the art can set the preset threshold by themselves. The smaller the value of the preset threshold, the better the optimization effect of the parameters of the AE-ICA network based on the autoencoder.

[0077] For example, the first time: if the difference between the original independent signal recovered for the first time and the original independent signal to be aliased is greater than or equal to a preset threshold, the parameters of the autoencoder-based AE-ICA network are optimized; the second time: based on the autoencoder-based AE-ICA network optimized for the first time, the original independent signal recovered for the second time (that is, the original independent signal recovered corresponding to the current iteration) is obtained. If the difference between the original independent signal recovered for the second time and the original independent signal to be aliased is less than the preset threshold, the optimization of the parameters of the autoencoder-based AE-ICA network is stopped.

[0078] Specifically, for ease of understanding, the process of separating the aliased signal to be separated is described. The AE-ICA network is used to perform blind source separation on the input aliased signal to be separated to obtain the restored original independent signal, including:

[0079] Obtain the aliased signal to be separated; use the Fast-ICA network of the AE-ICA network based on the autoencoder to separate the aliased signal to be separated, and obtain the original independent signal to be restored; input the aliased signal to be separated into the Encoder network of the AE-ICA network based on the autoencoder to obtain the latent space representation of the aliased signal to be separated; fuse the latent space representation with the original independent signal to be restored to obtain a fused signal; input the fused signal into the Decoder network of the AE-ICA network based on the autoencoder to obtain the restored original independent signal. At this point, the separation of the aliased signal is completed. It should be noted that the aliased signal to be separated input into the Fast-ICA network of the AE-ICA network based on the autoencoder can be preprocessed by decentering and whitening, and the aliased signal to be separated input into the Encoder network of the AE-ICA network based on the autoencoder is not preprocessed (i.e., no decentering and whitening preprocessing is performed).

[0080] Experiments and their results:

[0081] This paper uses four common modulation modes (FM, 4ASK, OOK, 64QAM) in the public data set RadioML 2018.01A for experimental demonstration. In the data set, a dual-channel IQ orthogonal modulation signal with a signal length of 1024 is used as the basic signal format of the network, and its basic signal size is shape(x)=[2,1024]. In order to facilitate network processing, the dual-channel signal is separated and spliced ​​to obtain a new signal size shape(x')=[1,2048].

[0082] The two original modulation signals (original modulation signal 1 and original modulation signal 2) are as follows Figure 4 As shown, 401 is the original modulation signal 1, and 402 is the original modulation signal 2. Select any two modulation modes from FM, 4ASK, OOK, and 64QAM to alias the two original modulation signals. The result of the first aliasing is as follows Figure 5 As shown, 501 is the signal after the original modulation signal 1 is modulated for the first time, and 502 is the signal after the original modulation signal 2 is modulated for the first time; the result of the second aliasing is as follows Figure 6 As shown, 601 is a signal after the original modulated signal 1 is modulated for the second time, and 602 is a signal after the original modulated signal 2 is modulated for the second time. The two modulation modes corresponding to the first aliasing are different from the two modulation modes corresponding to the second aliasing.

[0083] The original independent signals separated by the Fast-ICA algorithm are compared with the original independent signals separated by the AE-ICA algorithm of the present invention (i.e., the trained AE-ICA network based on the autoencoder).

[0084] like Figure 7 As shown, Figure 7 The figure shows an original modulated signal 1, an original independent signal 701 corresponding to the original modulated signal 1 separated by the Fast-ICA algorithm, and an original independent signal 702 corresponding to the original modulated signal 1 separated by the AE-ICA algorithm of the present invention.

[0085] like Figure 8 As shown, Figure 8 The figure shows the original modulated signal 2, the original independent signal 801 corresponding to the original modulated signal 2 separated by the Fast-ICA algorithm, and the original independent signal 802 corresponding to the original modulated signal 2 separated by the AE-ICA algorithm of the present invention.

[0086] This further illustrates that there is a large difference between the amplitude of the original independent signal separated by the Fast-ICA algorithm and the amplitude of the original modulated signal, and that the amplitude of the original independent signal separated by the AE-ICA algorithm of the present invention is closer to the amplitude of the original modulated signal.

[0087] For the phase inversion problem of the original independent signals separated by the Fast-ICA algorithm: Fig. 9 As shown, Fig. 9 The original modulation signal 3, the original independent signal 902 corresponding to the original modulation signal 3 separated by the Fast-ICA algorithm, and the original independent signal 903 corresponding to the original modulation signal 3 separated by the AE-ICA algorithm of the present invention are shown in FIG. 901 is the original modulation signal 3. It should be noted that, for example, the original modulation signal 3 is first aliased with another original modulation signal to obtain two aliased signals, and the two aliased signals are separated by the corresponding algorithm to obtain the original independent signal corresponding to the original modulation signal 3.

[0088] like Fig. 9 As shown, the phase of the original independent signal corresponding to the original modulated signal 3 separated by the Fast-ICA algorithm is reversed compared to the phase of the original modulated signal 3; however, the difference between the phase of the original independent signal corresponding to the original modulated signal 3 separated by the AE-ICA algorithm of the present invention and the phase of the original modulated signal 3 is small.

[0089] At the same time, a comprehensive comparison between the Fast-ICA algorithm and the AE-ICA algorithm of the present invention is also conducted on the test set (a total of 2358 signal pairs). Specifically, the mean of the L2 distance between the separation results of the 2358 signal pairs obtained using the Fast-ICA algorithm (i.e., the original independent signals after separation) and the corresponding original independent signals to be aliased (i.e., the signals before aliasing) is calculated, and the mean of the L2 distance between the separation results of the 2358 signal pairs obtained using the AE-ICA algorithm of the present invention and the corresponding original independent signals to be aliased is calculated, as shown in Table 1. Among them, the signal pair includes signal 1 and signal 2, and the L2 distance is the Euclidean distance.

[0090] Table 1

[0091]

[0092] It can be seen from Table 1 that the quality of the signal separated by the AE-ICA algorithm of the present invention is better than that of the Fast-ICA algorithm (has been significantly improved).

[0093] In summary, the experimental results show that the present invention can more accurately separate the original independent signal from the aliased signal, which is of great significance for speech or signal recognition.

[0094] An embodiment of the present application also provides a blind source separation device, including: a memory and a processor, the memory and the processor are connected via a bus communication, a computer program is stored in the memory, and the computer program can be run on the processor, thereby implementing the steps of the blind source separation method based on autoencoder combined with fast independent component analysis disclosed in the embodiment of the present application.

[0095] An embodiment of the present application also provides a computer-readable storage medium, on which a computer program / instruction is stored. When the computer program / instruction is executed by a processor, the steps of the blind source separation method based on autoencoder combined with fast independent component analysis as disclosed in the embodiment of the present application are implemented.

[0096] An embodiment of the present application also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the blind source separation method based on autoencoder combined with fast independent component analysis as disclosed in the embodiment of the present application.

[0097] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0098] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, devices or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0099] The embodiments of the present application are described with reference to the flowcharts and / or block diagrams of the methods, devices, storage media, and program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the 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 terminal device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the processes 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.

[0100] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0101] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A blind source separation method based on autoencoder combined with fast independent component analysis, characterized in that: include: Build and train an autoencoder-based AE-ICA network; The AE-ICA network is used to perform blind source separation on the input aliased signal to be separated, and the restored original independent signal is obtained; The AE-ICA network based on the autoencoder includes an Encoder network, a Fast-ICA network and a Decoder network; The method of training the autoencoder-based AE-ICA network includes: (1) generating an aliased signal based on the original independent signal to be aliased; (2) inputting the aliased signal into the Fast-ICA network for blind source separation to obtain the original independent signal to be restored; (3) inputting the aliased signal into the Encoder network to obtain a latent space representation of the aliased signal; (4) fusing the latent space representation with the original independent signal to be restored, obtaining a fused signal and inputting it into the Decoder network to obtain the restored original independent signal; (5) calculating the loss of the restored original independent signal and the original independent signal to be aliased and performing back propagation to optimize the parameters of the autoencoder-based AE-ICA network; (6) Iteratively optimizing the parameters of the autoencoder-based AE-ICA network until the number of iterations reaches a preset number and then stopping the iteration, or determining whether to continue iteratively optimizing the parameters of the autoencoder-based AE-ICA network based on the difference between the restored original independent signal corresponding to the current iteration and the original independent signal to be aliased.

2. The blind source separation method based on autoencoder combined with fast independent component analysis according to claim 1 is characterized in that: The Encoder network includes a first LSTM network, a linear layer Linear, an activation function ReLU and a linear layer Linear connected in sequence; The first LSTM network includes 3 layers of LSTM, the input dimension is the number of receiving channels 2, and the hidden layer is 256; The Decoder network includes a linear layer Linear, an activation function ReLU, a linear layer Linear, a second LSTM network and a linear layer Linear connected in sequence; The second LSTM network includes 3 layers of LSTM, the input dimension is 2, the number of receiving channels, and the hidden layer is 256.

3. The blind source separation method based on autoencoder combined with fast independent component analysis according to claim 1 or 2, characterized in that: Before the aliased signal is input into the Fast-ICA network, the method further includes: The aliased signal is decentralized so that the mean value of the processed signal is 0.

4. The blind source separation method based on autoencoder combined with fast independent component analysis according to claim 1 or 2, characterized in that: Before the Fast-ICA network performs blind source separation on the aliased signal, the method further includes: The aliased signal is pre-processed by "whitening".

5. The blind source separation method based on autoencoder combined with fast independent component analysis according to claim 1 or 2, characterized in that: The fusion of the latent space representation and the original independent signal to be restored is a dot product fusion.

6. The blind source separation method based on autoencoder combined with fast independent component analysis according to claim 1 or 2, characterized in that: The calculation formula for the loss of the restored original independent signal and the original independent signal to be aliased is: in, Represents the loss of the restored original independent signal and the original independent signal to be aliased, represents the original independent signal after recovery, S represents the original independent signal to be aliased, It is expressed as a vector 2-norm, n represents the total number of original independent signals to be aliased, and i represents the index of the original independent signal.

7. The blind source separation method based on autoencoder combined with fast independent component analysis according to claim 1 or 2, characterized in that: Determining whether to continue iteratively optimizing the parameters of the autoencoder-based AE-ICA network according to the difference between the restored original independent signal corresponding to the current iteration and the original independent signal to be aliased, including: In response to the difference between the restored original independent signal corresponding to the current iteration and the original independent signal to be aliased being less than a preset threshold, the iterative optimization of the parameters of the autoencoder-based AE-ICA network is stopped.

8. A blind source separation device, comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the blind source separation method based on autoencoder combined with fast independent component analysis described in any one of claims 1 to 7 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the blind source separation method based on autoencoder combined with fast independent component analysis according to any one of claims 1 to 7 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by a processor, the steps of the blind source separation method based on autoencoder combined with fast independent component analysis described in any one of claims 1 to 7 are implemented.

Images