Small-sample single-channel source separation anti-interference method and device based on multi-task learning

Abstract

The invention discloses a small-sample single-channel source separation anti-interference method and device based on multi-task learning. The method comprises the following steps: in a training stage, firstly, creating a multi-task training data set which comprises a target communication signal and a plurality of interference signals; then, building a signal separation network, and creating a joint training objective function; finally, inputting the multi-task signal into the network for training, and storing an optimal pre-training model; in the fine tuning stage, firstly, a fine tuning data set and a target function are created; then inputting the small sample signals into a pre-training model for fine tuning, and storing an optimal model; in the application stage, firstly, an interfered aliasing signal is obtained, the fine-tuned network model is input for signal separation, and then bit data of the separated signal is recovered. The device comprises a multi-task training module, a small sample fine tuning module and an application module. According to the invention, communication signals and interference signals can be effectively separated under the condition of single antenna reception, and the anti-interference capability of wireless communication is improved.

Description

Technical Field

[0001] The present invention relates to the field of signal processing technology, and in particular to a small sample single-channel source separation anti-interference method and device based on multi-task learning. Background Art

[0002] With the rapid development of wireless communication technology, the stability and reliability of communication links have become the key to information transmission. However, communication signals will inevitably be interfered by natural noise, technical equipment, and even malicious interference attacks during transmission. These interferences seriously affect the quality of communication and may even cause communication failure. Therefore, studying how to recover useful signals from various interferences has become a core topic in the field of communication anti-interference.

[0003] Traditional anti-interference technologies, such as frequency hopping spread spectrum, direct sequence spread spectrum, adaptive filtering and beamforming, can identify and suppress interference to a certain extent, but in complex and changeable interference environments, especially when facing unknown or new interference, it is often difficult to achieve ideal results. In addition, traditional methods are difficult to effectively formulate interference feature analysis and anti-interference strategies when there are only a small number of interference samples. Specifically, frequency hopping spread spectrum relies on frequency hopping patterns to avoid interference, but it may be difficult to avoid when facing broadband interference; direct sequence spread spectrum resists interference by expanding signal energy to a wider frequency range, but it is more suitable for scenarios with high reliability requirements; adaptive filtering technology needs to continuously adjust its parameters according to the characteristics of the received signal. In the face of unknown interference, the filter may take a long time to converge to the appropriate parameter settings, during which the communication quality may be affected, and if the interference characteristics change rapidly, the filter may find it difficult to keep up with such changes; beamforming technology uses multi-antenna technology to transmit and receive signals in a directional manner to enhance signals and suppress interference, but it is not applicable in single-channel scenarios.

[0004] Blind source separation attempts to recover independent sources from mixed signals and is an effective means of separating co-channel interference. Traditional blind source separation methods include ICA and NMF, but such methods usually use multi-channel separation, and single-channel separation is not effective. Single-channel blind source separation technology has always been at the forefront of research in the field of signal processing. It aims to use only a single receiving antenna to distinguish and recover multiple source signals mixed together under resource-constrained conditions. This technology has significant advantages, such as reducing the required hardware equipment and reducing costs, but its challenges are equally obvious, especially in environments with low signal-to-noise ratios. Traditional algorithms often find it difficult to provide satisfactory separation effects, and are accompanied by a huge computational burden.

[0005] In order to overcome these problems, deep learning technology has been introduced into the research of single-channel blind source separation. Deep learning methods have made breakthrough progress in speech signal separation by using their powerful feature extraction and representation learning capabilities. These methods can be divided into two categories: one is the method that operates in the time-frequency domain, such as deep clustering learning, permutation invariant training, etc., which captures the frequency components of the signal by converting the speech signal from the time domain to the time-frequency domain; the other is the method that processes directly in the time domain, such as using long short-term memory (LSTM) recurrent networks and convolutional recurrent networks (CRN), which can process time series data and learn the temporal dynamic characteristics of the signal. With the rapid development of these speech separation technologies, their application in the field of wireless communications has also shown great potential. Deep learning models can not only handle complex signal mixing problems, but also adapt to changing communication environments, thereby improving the accuracy and robustness of signal separation. In addition, the adaptive learning ability of deep learning models enables them to maintain high separation performance when facing unknown or dynamically changing signal sources.

[0006] Although the application of deep learning in the field of single-channel blind source separation is full of potential, it also faces some challenges, especially in anti-interference communication scenarios. Since it is difficult to obtain enough interference samples for newly emerging interference types, there are often only a small number of samples available to train the model, which limits the learning and adaptability of deep learning models.

[0007] One of the core goals of deep learning is to imitate the human perceptual system. The human visual system can quickly recognize new categories with very few labeled samples. This ability has inspired the rapid development of small sample transfer learning in the field of image classification. Research in this field has made significant progress, involving a variety of transfer learning strategies, including initialization parameters, distance metrics, and domain adaptation. Among the initialization parameter strategies, multi-task learning (MTL) has attracted much attention due to its potential in improving model generalization ability and efficiency. Multi-task learning improves model performance through learning strategies such as hard parameter sharing and soft parameter sharing. Summary of the invention

[0008] The purpose of the present invention is to provide a small sample single-channel source separation and anti-interference method and device that can effectively separate communication signals and interference signals when a single antenna is received, effectively deal with interference situations of unknown or small samples, significantly improve the generalization ability of the separation model, and thus improve the anti-interference ability of wireless communication.

[0009] The technical solution to achieve the purpose of the present invention is: a small sample single-channel source separation anti-interference method based on multi-task learning, including a multi-task training stage, a fine-tuning stage and an application stage, wherein:

[0010] The multi-task training phase includes the following steps:

[0011] Step A1, creating a training data set, including a communication signal, a plurality of known interference signals, and a mixed signal of the two;

[0012] Step A2: Building a signal separation network, including encoder, separator and decoder modules;

[0013] Step A3: Create a joint training objective function for multi-task learning;

[0014] Step A4: Based on the joint training objective function, the signal data set obtained in step A1 is used to train the signal separation network in step A2 to obtain a trained pre-trained signal separation network;

[0015] The small sample fine-tuning phase includes the following steps:

[0016] Step B1, creating a fine-tuning data set, including a communication signal, a small sample target interference signal, and a mixed signal of the two;

[0017] Step B2, creating a fine-tuning phase objective function;

[0018] Step B3: Based on the objective function of the fine-tuning stage, the pre-trained signal separation network obtained in step A4 is fine-tuned using the fine-tuning data set obtained in step B1 to obtain a fine-tuned signal separation network;

[0019] The application phase includes the following steps:

[0020] Step C1, obtaining a small sample target interference aliasing signal to be separated through a single antenna;

[0021] Step C2, using the signal separation network fine-tuned in step B3, separating the target communication signal and the interference signal from the aliased signal, and completing the blind separation of the aliased signal;

[0022] Step C3: Send the separated signal obtained in step C2 to a demodulator for demodulation to restore the bit data of the separated signal.

[0023] Furthermore, the creation of the training data set described in step A1 is performed in any of the following two ways:

[0024] The first method: using computer software to generate communication signals and interference signals of various modulation modes, the modulation mode is digital modulation or analog modulation, and the interference signals include digital modulation interference signals, analog modulation interference signals, noise frequency modulation interference signals, noise amplitude modulation interference signals, single-tone interference signals, multi-tone interference signals, comb spectrum interference signals, broadband noise interference, and swept frequency interference;

[0025] Based on the generated target communication signal and interference signal, the target communication signal and multiple interference signals are randomly mixed to form an aliased signal, the interference signal ratio ISR is set to 0-8dB, and noise is added to the aliased signal to generate an aliased communication signal, and the signal-to-noise ratio SNR is -5dB to 20dB;

[0026] The aliased communication signal is shown as follows:

[0027]

[0028] Where x(t) is the time-frequency mixed communication signal, s i (t)∈C 1×T represents the source signal, T represents the signal length, i=1,2,...,λ, i represents the i-th source signal, λ represents the total number of source signals, a i represents the unknown mixing coefficient corresponding to the i-th source signal, and n(t) represents Gaussian white noise;

[0029] The multi-task training data set contains the target communication signal and multiple interference signals, and the target communication signal and each interference signal are mixed separately;

[0030] The second method is to sample actual communication equipment and interference equipment to obtain data sets of target communication signals, interference signals, and aliased signals.

[0031] Furthermore, the signal separation network described in step A2 is constructed as follows:

[0032] The goal of the single-channel source separation task is to extract the mixed signal x∈R T Separate N source signals s1, s2, ..., s N ∈R T ; A modified separation transformer, namely SepFormer model, is used, which has a Transformer-based architecture without recurrent neural network RNN, including encoder, decoder and separator. The encoder and separator modules are modified, including the configuration of convolutional layers and activation functions, and the convolution kernel and step size are reduced during the convolution process, as follows:

[0033] (1) Encoder

[0034] The encoder adopts a bottleneck design, replacing the one-dimensional convolution, through a double convolution layer, initially increasing and then reducing the number of kernels to accurately extract the coding features of the signal; the encoder receives the time domain mixed signal x and generates the learned two-dimensional feature representation h∈R F×T′ , where F is the feature dimension, T′ is the new length of the signal, and a convolutional layer is used: h = conv1d(ReLU(conv1d(x))), where conv1d() represents a one-dimensional convolution and ReLU() represents a nonlinear activation function;

[0035] (2) Separator

[0036] The separator receives the encoded 2D feature representation h and estimates a mask m1,...,m for each source in the mixture. N , that is: m k =separator(h), k=1,...,N; where separator() represents the separator, and N represents the number of signal sources;

[0037] In the process of estimating the mask, the separator first processes the 2D feature representation h into overlapping 3D feature blocks h′ and inputs h′ into the separation transform block: Where F represents the feature dimension, C is the length of each block, and T c is the number of blocks;

[0038] The separation transformer block is the core of the separator. It uses a dual-scale approach to simulate short-term and long-term dependencies. It consists of two Transformers: the inner loop Transformer, i.e., IntraT, and the outer loop Transformer, i.e., InterT. IntraT simulates short-term dependencies within each block, and InterT simulates long-term dependencies across blocks after permuting the dimensions.

[0039] The overall transformation of the separated transform block is defined as:

[0040]

[0041] Among them, f intra and f inter They represent IntraT and InterT transformations, respectively, P represents the permutation operation, h′ represents the three-dimensional feature block, and h″ represents the feature after separation transformation;

[0042] (3) Decoder

[0043] The decoder uses the transposed convolutional layers and the masks predicted by the mask network to reconstruct the separated signals in the time domain:

[0044]

[0045] in, represents the separated k-th source, conv1d-transpose() represents one-dimensional inverse convolution, m k Represents the mask of the k-th source.

[0046] Furthermore, the joint training objective function for creating multi-task learning in step A3 is as follows:

[0047] During the training phase, the parameters of the encoder, decoder, and separator are shared, and a unified objective function is used to minimize the combined loss of all tasks, namely:

[0048]

[0049] Where J(θ) represents the total loss of the training task, θ represents the parameter set shared by all tasks, and α j is the loss L assigned to the jth task j If all tasks are equally important, then α j =1 / M;

[0050] The parameters of the separation model are b Optimize across multiple tasks in an alternating manner, namely:

[0051]

[0052] Among them, β represents the learning rate, Indicates gradient descent along θ;

[0053] The loss L for each task i The scale-invariant signal-to-noise ratio SI-SNR of the sentence-level permutation-invariant uPIT loss is used as the loss function, and the scale-invariant signal-to-noise ratio SI-SNR is expressed as:

[0054]

[0055] Among them, s k and represents the source signal and the estimated signal, <·,·> represents the dot product of two vectors, ||·||2 represents the two-norm, s target represents the projection component of the estimated signal in the direction of the source signal, e noise Represents the vertical component of the estimated signal in the direction of the source signal.

[0056] Furthermore, based on the joint training objective function described in step A4, the signal separation network of step A2 is trained using the signal data set obtained in step A1 to obtain a trained pre-trained signal separation network, as follows:

[0057] The data of the multi-task training set obtained in step A1 is packaged, the sequences are combined into batches, and sent to the signal separation network in step A2 for training. The network parameters are set, and the Adam optimizer is used with an initial learning rate of β = 0.0125. A total of 100 rounds of training are conducted. After the training is completed, the optimal parameter model is saved.

[0058] Furthermore, the creation of the fine-tuning data set described in step B1 includes the communication signal and the small sample target interference signal and a mixed signal of the two, as follows:

[0059] The target communication signal and the small sample target interference signal are obtained by either of two methods:

[0060] The first method: computer software is used to generate communication signals and small sample interference signals. The communication signal modulation method is digital modulation or analog modulation. The interference signals include digital modulation interference signals, analog modulation interference signals, noise frequency modulation interference signals, noise amplitude modulation interference signals, single-tone interference signals, multi-tone interference signals, comb spectrum interference signals, broadband noise interference, and swept frequency interference.

[0061] Based on the generated target communication signal and interference signal, the target communication signal and the target interference signal are randomly mixed to form an aliased signal, and noise is added to the aliased signal to generate an aliased communication signal with a signal-to-noise ratio (SNR) of -5dB to 20dB;

[0062] The second method is to sample the actual communication equipment and the interference equipment to obtain the data sets of the target communication signal, the target interference signal and the aliased signal;

[0063] The acquired target communication signal is the same as the communication signal in step A1, the interference signal is of a different type from the interference signal in step A1, and the number of small samples is 0-10.

[0064] Furthermore, the objective function of the fine-tuning phase described in step B2 is created as follows:

[0065] SI-SNR with uPIT loss is adopted as the fine-tuning loss function.

[0066] Furthermore, based on the objective function of the fine-tuning stage described in step B3, the pre-trained signal separation network obtained in step A4 is fine-tuned using the fine-tuning data set obtained in step B1 to obtain a fine-tuned signal separation network, which is specifically as follows:

[0067] Use a small sample data set to fine-tune the pre-trained model. The fine-tuning rounds are set to 20 rounds. During the fine-tuning process, choose to freeze the encoder, decoder, or separator of the pre-trained model. After fine-tuning, save the optimal model.

[0068] A small sample single-channel source separation anti-interference device based on multi-task learning, the device is used to implement the small sample single-channel source separation anti-interference method based on multi-task learning, the device includes a multi-task training module, a small sample fine-tuning module and an application module, wherein:

[0069] The multi-task training module includes a data set creation unit, a signal separation network construction unit, an objective function creation unit, and a network training unit, wherein:

[0070] A data set creation unit, which creates a multi-task training data set, including communication signals and various interference signals;

[0071] Signal separation network construction unit, building encoder, separator and decoder modules;

[0072] Joint training objective function creation unit, creating a customized multi-task learning training objective function;

[0073] The network training unit uses the signal data set obtained by the data set creation unit to train the network obtained by the signal separation network construction unit based on the joint training objective function, and obtains a trained pre-trained separation network;

[0074] The small sample fine-tuning module includes a small sample data set creation unit, a fine-tuning objective function creation unit and a fine-tuning network unit, wherein:

[0075] A small sample data set creation unit, which creates a fine-tuning data set, including a communication signal and a small sample target interference signal and a mixed signal of the two;

[0076] A fine-tuning objective function creation unit is used to create an objective function in the fine-tuning stage;

[0077] Fine-tune the network unit, based on the objective function of the fine-tuning stage, and use the fine-tuning dataset to fine-tune the trained pre-trained separation model to obtain a fine-tuned separation network;

[0078] The application module includes a signal acquisition unit, a signal separation unit, and a demodulation unit, wherein:

[0079] A signal acquisition unit, which acquires the time-frequency aliased signal to be separated through a single antenna;

[0080] The signal separation unit uses a fine-tuned separation network to separate the target communication signal and the interference signal from the aliased signal, thereby completing the separation of the aliased signal;

[0081] The demodulation unit sends the separated communication signal obtained by the signal separation unit to the demodulator for demodulation to restore the bit data of the signal.

[0082] A mobile terminal comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the small sample single-channel source separation and anti-interference method based on multi-task learning is implemented.

[0083] Compared with the prior art, the present invention has the following significant advantages: (1) It adopts a strategy of combining multi-task learning with small sample transfer learning, builds a model that can handle multiple types of interference at the same time, takes advantage of multi-task learning, shares knowledge to improve the generalization ability of the model, and solves the communication anti-interference problem when the number of target interference samples is insufficient; (2) It adopts a separation model as the basic framework, and through a multi-task learning pre-training method, guides the model to learn how to identify and adapt to various interferences, improves the robustness and flexibility of the model in practical applications, can effectively deal with interference situations of unknown or small samples, and significantly improves the generalization ability of the separation model. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] Figure 1 It is a flow chart of the small sample single-channel source separation and anti-interference method based on multi-task learning of the present invention.

[0085] Figure 2 Schematic diagram of a scenario of a small sample single-channel source separation and anti-interference method based on multi-task learning in an embodiment.

[0086] Figure 3 Schematic diagram of the structure of multi-task learning training and fine-tuning network in an embodiment of the present invention.

[0087] Figure 4 Schematic diagram of the structure of the separation network used in multi-task learning in an embodiment of the present invention. DETAILED DESCRIPTION

[0088] It is easy to understand that according to the technical solution of the present invention, without changing the essential spirit of the present invention, a person skilled in the art can imagine various implementation methods of the present invention. Therefore, the following specific implementation methods and drawings are only exemplary descriptions of the technical solution of the present invention, and should not be regarded as the whole of the present invention or as a limitation or restriction of the technical solution of the present invention.

[0089] Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that the relative arrangement of components and steps, numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present invention unless otherwise specifically stated.

[0090] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the invention, its application, or uses.

[0091] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered as part of the specification.

[0092] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not limiting. Therefore, other examples of the exemplary embodiments may have different values.

[0093] Combination Figure 1 The present invention provides a small sample single-channel source separation anti-interference method based on multi-task learning, including a multi-task training stage, a fine-tuning stage and an application stage, wherein:

[0094] The multi-task training phase includes the following steps:

[0095] Step A1, creating a training data set, including a communication signal, a plurality of known interference signals, and a mixed signal of the two;

[0096] Step A2: Building a signal separation network, including encoder, separator and decoder modules;

[0097] Step A3: Create a joint training objective function for multi-task learning;

[0098] Step A4: Based on the joint training objective function, the signal data set obtained in step A1 is used to train the signal separation network in step A2 to obtain a trained pre-trained signal separation network;

[0099] The small sample fine-tuning phase includes the following steps:

[0100] Step B1, creating a fine-tuning data set, including a communication signal, a small sample target interference signal, and a mixed signal of the two;

[0101] Step B2, creating a fine-tuning phase objective function;

[0102] Step B3: Based on the objective function of the fine-tuning stage, the pre-trained signal separation network obtained in step A4 is fine-tuned using the fine-tuning data set obtained in step B1 to obtain a fine-tuned signal separation network;

[0103] The application phase includes the following steps:

[0104] Step C1, obtaining a small sample target interference aliasing signal to be separated through a single antenna;

[0105] Step C2, using the signal separation network fine-tuned in step B3, separating the target communication signal and the interference signal from the aliased signal, and completing the blind separation of the aliased signal;

[0106] Step C3: Send the separated signal obtained in step C2 to a demodulator for demodulation to restore the bit data of the separated signal.

[0107] As a specific example, the creation of the training data set described in step A1 is performed in any of the following two ways:

[0108] The first method: using computer software to generate communication signals and interference signals of various modulation modes, the modulation mode is digital modulation or analog modulation, and the interference signals include digital modulation interference signals, analog modulation interference signals, noise frequency modulation interference signals, noise amplitude modulation interference signals, single-tone interference signals, multi-tone interference signals, comb spectrum interference signals, broadband noise interference, and swept frequency interference;

[0109] Based on the generated target communication signal and interference signal, the target communication signal and multiple interference signals are randomly mixed to form an aliased signal, the interference signal ratio ISR is set to 0-8dB, and noise is added to the aliased signal to generate an aliased communication signal, and the signal-to-noise ratio SNR is -5dB to 20dB;

[0110] The aliased communication signal is shown as follows:

[0111]

[0112] Where x(t) is the time-frequency mixed communication signal, s i (t)∈C 1×T represents the source signal, T represents the signal length, i=1,2,...,λ, i represents the i-th source signal, λ represents the total number of source signals, a i represents the unknown mixing coefficient corresponding to the i-th source signal, and n(t) represents Gaussian white noise;

[0113] The multi-task training data set contains the target communication signal and multiple interference signals, and the target communication signal and each interference signal are mixed separately;

[0114] The second method is to sample actual communication equipment and interference equipment to obtain data sets of target communication signals, interference signals, and aliased signals.

[0115] As a specific example, the signal separation network constructed in step A2 is as follows:

[0116] The signal separation network uses known classic networks such as Conv-TasNet and SepFormer, or uses a self-built network model;

[0117] The goal of the single-channel source separation task is to extract the mixed signal x∈R T Separate N source signals s1, s2, ..., s N ∈R T; The present invention adopts a modified separation transformer, namely SepFormer model, which is a classic speech separation network. The model has a Transformer-based architecture without a recurrent neural network (RNN), including an encoder, a decoder and a separator; in order to better adapt the network to the characteristics of the communication signal, the encoder and separator modules are modified, including the configuration of the convolution layer and the activation function, including the configuration of the convolution layer and the activation function; due to the low information packing density of the communication signal, a smaller convolution kernel and step size are implemented in the convolution process, as follows:

[0118] (1) Encoder

[0119] The encoder adopts a bottleneck design, replacing the previous one-dimensional convolution. Through the double convolution layer, the number of kernels is initially increased and then reduced, so as to more accurately extract the coding features of the signal; the encoder receives the time domain mixed signal x and generates the learned two-dimensional feature representation h∈R F×T′ , where F is the feature dimension, T′ is the new length of the signal, and a convolutional layer is used: h = conv1d(ReLU(conv1d(x))), where conv1d() represents a one-dimensional convolution and ReLU() represents a nonlinear activation function;

[0120] (2) Separator

[0121] The separator receives the encoded 2D feature representation h and estimates a mask m1,...,m for each source in the mixture. N , that is: m k =separator(h), k=1,...,N; where separator() represents the separator, and N represents the number of signal sources;

[0122] In the process of estimating the mask, the separator first processes the 2D feature representation h into overlapping 3D feature blocks h′ and inputs h′ into the separation transform block: Where F represents the feature dimension, C is the length of each block, and T c is the number of blocks;

[0123] The separation transformer block is the core of the separator. It uses a dual-scale approach to simulate short-term and long-term dependencies. It consists of two Transformers: the inner loop Transformer, i.e., IntraT, and the outer loop Transformer, i.e., InterT. IntraT simulates short-term dependencies within each block, and InterT simulates long-term dependencies across blocks after permuting the dimensions.

[0124] The overall transformation of the separated transform block is defined as:

[0125]

[0126] Among them, f intra and f inter They represent IntraT and InterT transformations, respectively, P represents the permutation operation, h′ represents the three-dimensional feature block, and h″ represents the feature after separation transformation;

[0127] (3) Decoder

[0128] The decoder uses the transposed convolutional layers and the masks predicted by the mask network to reconstruct the separated signals in the time domain:

[0129]

[0130] in, represents the separated k-th source, conv1d-transpose() represents one-dimensional inverse convolution, m k Represents the mask of the k-th source.

[0131] As a specific example, the joint training objective function for creating multi-task learning in step A3 is as follows:

[0132] During the training phase, the parameters of the encoder, decoder, and separator are shared, and a unified objective function is used to minimize the combined loss of all tasks, namely:

[0133]

[0134] where J(θ) represents the total loss of the training task, θ represents the set of parameters shared by all tasks; due to the possibility that some tasks may be more important than others, α is used j Denotes the loss L assigned to the jth task j If all tasks are equally important, then α j =1 / M;

[0135] The parameters of the separation model are b Optimize across multiple tasks in an alternating manner, namely:

[0136]

[0137] Among them, β represents the learning rate, Indicates gradient descent along θ;

[0138] The loss L for each task i The scale-invariant signal-to-noise ratio SI-SNR with sentence-level permutation-invariant uPIT loss is used as the loss function;

[0139] The scale-invariant signal-to-noise ratio SI-SNR is expressed as:

[0140]

[0141] Among them, s k and represents the source signal and the estimated signal, <·,·> represents the dot product of two vectors, ||·||2 represents the two-norm, s target represents the projection component of the estimated signal in the direction of the source signal, e noise Represents the vertical component of the estimated signal in the direction of the source signal.

[0142] As a specific example, based on the joint training objective function described in step A4, the signal separation network of step A2 is trained using the signal data set obtained in step A1 to obtain a trained pre-trained signal separation network, as follows:

[0143] The data of the multi-task training set obtained in step A1 is packaged, the sequences are combined into batches, and sent to the signal separation network in step A2 for training. The network parameters are set, and the Adam optimizer is used with an initial learning rate of β = 0.0125. A total of 100 rounds of training are conducted. After the training is completed, the optimal parameter model is saved.

[0144] As a specific example, the creation of the fine-tuning data set described in step B1 includes a communication signal and a small sample target interference signal and a mixed signal of the two, as follows:

[0145] The target communication signal and the small sample target interference signal are obtained by either of two methods:

[0146] The first method: computer software is used to generate communication signals and small sample interference signals. The communication signal modulation method is digital modulation or analog modulation. The interference signals include digital modulation interference signals, analog modulation interference signals, noise frequency modulation interference signals, noise amplitude modulation interference signals, single-tone interference signals, multi-tone interference signals, comb spectrum interference signals, broadband noise interference, and swept frequency interference.

[0147] Based on the generated target communication signal and interference signal, the target communication signal and the target interference signal are randomly mixed to form an aliased signal, and noise is added to the aliased signal to generate an aliased communication signal with a signal-to-noise ratio (SNR) of -5dB to 20dB;

[0148] The second method is to sample the actual communication equipment and the interference equipment to obtain the data sets of the target communication signal, the target interference signal and the aliased signal;

[0149] The acquired target communication signal is the same as the communication signal in step A1, the interference signal is of a different type from the interference signal in step A1, and the number of small samples is 0-10.

[0150] As a specific example, the objective function of the fine-tuning phase described in step B2 is created as follows:

[0151] SI-SNR with uPIT loss is adopted as the fine-tuning loss function.

[0152] As a specific example, based on the fine-tuning phase objective function described in step B3, the trained pre-trained signal separation network obtained in step A4 is fine-tuned using the fine-tuning data set obtained in step B1 to obtain a fine-tuned signal separation network, which is specifically as follows:

[0153] Use a small sample dataset to fine-tune the pre-trained model. The number of fine-tuning rounds is set to 20. During the fine-tuning process, you can choose to freeze some layers of the pre-trained model, such as freezing the encoder, decoder, or separator. After fine-tuning, save the optimal model.

[0154] The present invention also provides a small sample single-channel source separation anti-interference device based on multi-task learning, which is used to implement a small sample single-channel source separation anti-interference method based on multi-task learning. The device includes a multi-task training module, a small sample fine-tuning module and an application module, wherein:

[0155] The multi-task training module includes a data set creation unit, a signal separation network construction unit, an objective function creation unit, and a network training unit, wherein:

[0156] A data set creation unit, which creates a multi-task training data set, including communication signals and various interference signals;

[0157] Signal separation network construction unit, building encoder, separator and decoder modules;

[0158] Joint training objective function creation unit, creating a customized multi-task learning training objective function;

[0159] The network training unit uses the signal data set obtained by the data set creation unit to train the network obtained by the signal separation network construction unit based on the joint training objective function, and obtains a trained pre-trained separation network;

[0160] The small sample fine-tuning module includes a small sample data set creation unit, a fine-tuning objective function creation unit and a fine-tuning network unit, wherein:

[0161] A small sample data set creation unit, which creates a fine-tuning data set, including a communication signal and a small sample target interference signal and a mixed signal of the two;

[0162] A fine-tuning objective function creation unit is used to create an objective function in the fine-tuning stage;

[0163] Fine-tune the network unit, based on the objective function of the fine-tuning stage, and use the fine-tuning dataset to fine-tune the trained pre-trained separation model to obtain a fine-tuned separation network;

[0164] The application module includes a signal acquisition unit, a signal separation unit, and a demodulation unit, wherein:

[0165] A signal acquisition unit, which acquires the time-frequency aliased signal to be separated through a single antenna;

[0166] The signal separation unit uses a fine-tuned separation network to separate the target communication signal and the interference signal from the aliased signal, thereby completing the separation of the aliased signal;

[0167] The demodulation unit sends the separated communication signal obtained by the signal separation unit to the demodulator for demodulation to restore the bit data of the signal.

[0168] The present invention also provides a mobile terminal, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the small sample single-channel source separation and anti-interference method based on multi-task learning is implemented.

[0169] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0170] Example

[0171] This embodiment provides a small sample single-channel source separation anti-interference method based on multi-task learning, including a multi-task training stage, a fine-tuning stage and an application stage, wherein:

[0172] The multi-task training phase includes the following steps:

[0173] Step A1, creating a training data set, including a communication signal, a plurality of known interference signals, and a mixed signal of the two;

[0174] Step A2: Building a signal separation network, including encoder, separator and decoder modules;

[0175] Step A3: Create a joint training objective function for multi-task learning;

[0176] Step A4: Based on the joint training objective function, the signal data set obtained in step A1 is used to train the signal separation network in step A2 to obtain a trained pre-trained signal separation network;

[0177] The small sample fine-tuning phase includes the following steps:

[0178] Step B1, creating a fine-tuning data set, including a communication signal, a small sample target interference signal, and a mixed signal of the two;

[0179] Step B2, creating a fine-tuning phase objective function;

[0180] Step B3: Based on the objective function of the fine-tuning stage, the pre-trained signal separation network obtained in step A4 is fine-tuned using the fine-tuning data set obtained in step B1 to obtain a fine-tuned signal separation network;

[0181] The application phase includes the following steps:

[0182] Step C1, obtaining a small sample target interference aliasing signal to be separated through a single antenna;

[0183] Step C2, using the signal separation network fine-tuned in step B3, separating the target communication signal and the interference signal from the aliased signal, and completing the blind separation of the aliased signal;

[0184] Step C3: Send the separated signal obtained in step C2 to a demodulator for demodulation to restore the bit data of the separated signal.

[0185] Figure 2 This is a scenario diagram of a small sample single-channel source separation anti-interference method based on multi-task learning, which aims to combat unknown interference. The transmitter and receiver in the figure are responsible for sending and receiving signals respectively, while the interference source simulates the interference that may be encountered during the communication process. The goal of source separation anti-interference is to separate the source signal and the interference signal from the received interfered signal to ensure normal communication. Due to the lack of training data for unknown interference, known interference sources are used for training. Through the multi-task learning method, the model learns how to identify and adapt to interference. Multi-task learning allows the model to handle multiple types of interference at the same time and improve generalization ability by sharing knowledge. Through this training strategy, the system can use existing knowledge and experience to achieve effective interference identification and suppression when facing unknown interference, ensuring the stability and reliability of communication.

[0186] Figure 3 Schematic diagram of the structure of multi-task learning training and fine-tuning network. In the multi-task training stage, create a dataset X b , including communication signals and multiple interference signals, based on the joint training objective function, the data set is used to train the model to obtain a trained pre-trained separation network, as follows:

[0187] The data of the multi-task training set is packaged, the sequences are combined into batches, and sent to the separation network. The network parameters are set, and the Adam optimizer is used with an initial learning rate of β = 0.0125. A total of 100 rounds of training are conducted. After the training is completed, the optimal parameter model is saved.

[0188] In the fine-tuning phase, create a fine-tuning dataset X n , including communication signals and small sample target interference signals and mixed signals of the two, target communication signals and Xb The communication signal in is the same as that in X. b The types of interference signals are different, and the number of small samples is 0-10.

[0189] Based on the objective function of the fine-tuning stage, the fine-tuning dataset is used to fine-tune the pre-trained separation model to obtain the fine-tuned separation network. The details are as follows: the pre-trained model is fine-tuned using a small sample dataset. The fine-tuning round is set to 20 rounds. Too many fine-tuning rounds are prone to overfitting. During the fine-tuning process, you can choose to freeze some layers of the pre-trained model, such as freezing the encoder, decoder or separator; after fine-tuning, save the optimal model.

[0190] Figure 4 This is a schematic diagram of the separation network used in multi-task learning. Since multi-task learning is a model-independent learning, the signal separation network can use known classic networks such as Conv-TasNet, SepFormer, etc., or a network model built by yourself.

[0191] In the single-channel source separation task, the goal of source separation anti-interference is to extract the mixed signal x∈R T Separate N source signals s1, s2, ..., s N ∈R T . A modified SepFormer model is used in this embodiment, which has a Transformer-based architecture without a recurrent neural network RNN, including an encoder, a decoder, and a separator. In order to better adapt the network to the characteristics of the communication signal, the encoder and separator modules are modified, including the configuration of the convolution layer and the activation function; due to the low information packing density of the communication signal, a smaller convolution kernel and step size are implemented in the convolution process; specifically as follows:

[0192] Encoder: The encoder adopts a bottleneck design, replacing the previous one-dimensional convolution. Through double convolution layers, the number of kernels is initially increased and then reduced, so as to more accurately extract the coding features of the signal; the encoder receives the time domain mixed signal x and generates the learned two-dimensional feature representation h∈R F×T′ , where F is the feature dimension and T′ is the new length of the signal, using a convolutional layer: h = conv1d(ReLU(conv1d(x)));

[0193] Separator: The separator receives the encoded 2D feature representation h and estimates a mask m1,...,m for each source in the mixture. N , that is: m k =separator(h),k=1,...,N;

[0194] In the process, it first processes the representation h into overlapping blocks h′ and inputs them into the SepFormer block: Where C is the length of each block, T c is the number of blocks;

[0195] The SepFormer block is the core of the separator and uses a dual-scale approach to model short-term and long-term dependencies. It consists of two Transformers, IntraT and InterT. IntraT models short-term dependencies within each block, and InterT models long-term dependencies across blocks after permuting the dimensions.

[0196] The overall transformation of the SepFormer block is defined as: where f intra and f inter They represent IntraT and InterT conversions respectively, and P represents a permutation operation;

[0197] Decoder: The decoder uses the transposed convolutional layer and the mask predicted by the mask network to reconstruct the separated signal in the time domain:

[0198]

[0199] in represents the separated k-th source.

[0200] In summary, the small sample single-channel source separation anti-interference method and device based on multi-task learning proposed in the present invention solves the communication anti-interference problem when the number of target interference samples is insufficient; simulation results show that the designed multi-task learning signal separation method can effectively deal with the interference of unknown or small samples, and significantly improves the generalization ability of the separation model.

[0201] The above are only preferred embodiments of the present invention. It should be pointed out that, for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A small sample single channel source separation anti-interference method based on multi-task learning, characterized in that: It includes multi-task training phase, fine-tuning phase and application phase, among which: The multi-task training phase includes the following steps: Step A1, creating a training data set, including a communication signal, a plurality of known interference signals, and a mixed signal of the two; Step A2: Building a signal separation network, including encoder, separator and decoder modules; Step A3: Create a joint training objective function for multi-task learning; Step A4: Based on the joint training objective function, the signal data set obtained in step A1 is used to train the signal separation network in step A2 to obtain a trained pre-trained signal separation network; The small sample fine-tuning phase includes the following steps: Step B1, creating a fine-tuning data set, including a communication signal, a small sample target interference signal, and a mixed signal of the two; Step B2, creating a fine-tuning phase objective function; Step B3: Based on the objective function of the fine-tuning stage, the pre-trained signal separation network obtained in step A4 is fine-tuned using the fine-tuning data set obtained in step B1 to obtain a fine-tuned signal separation network; The application phase includes the following steps: Step C1, obtaining a small sample target interference aliasing signal to be separated through a single antenna; Step C2, using the signal separation network fine-tuned in step B3, separating the target communication signal and the interference signal from the aliased signal, and completing the blind separation of the aliased signal; Step C3: Send the separated signal obtained in step C2 to a demodulator for demodulation to restore the bit data of the separated signal.

2. The small sample single channel source separation anti-interference method based on multi-task learning according to claim 1 is characterized in that: The training dataset described in step A1 is created in one of the following two ways: The first method: using computer software to generate communication signals and interference signals of various modulation modes, the modulation mode is digital modulation or analog modulation, and the interference signals include digital modulation interference signals, analog modulation interference signals, noise frequency modulation interference signals, noise amplitude modulation interference signals, single-tone interference signals, multi-tone interference signals, comb spectrum interference signals, broadband noise interference, and swept frequency interference; Based on the generated target communication signal and interference signal, the target communication signal and multiple interference signals are randomly mixed to form an aliased signal, the interference signal ratio ISR is set to 0-8dB, and noise is added to the aliased signal to generate an aliased communication signal, and the signal-to-noise ratio SNR is -5dB to 20dB; The aliased communication signal is shown below: Where x(t) is the time-frequency mixed communication signal, s i (t)∈C 1×T represents the source signal, T represents the signal length, i=1,2,...,λ, i represents the i-th source signal, λ represents the total number of source signals, a i represents the unknown mixing coefficient corresponding to the i-th source signal, and n(t) represents Gaussian white noise; The multi-task training data set contains the target communication signal and multiple interference signals, and the target communication signal and each interference signal are mixed separately; The second method is to sample actual communication equipment and interference equipment to obtain data sets of target communication signals, interference signals, and aliased signals.

3. The small sample single channel source separation anti-interference method based on multi-task learning according to claim 2 is characterized in that: The signal separation network described in step A2 is constructed as follows: The goal of the single-channel source separation task is to extract the mixed signal x∈R T Separate N source signals s1, s2, ..., s N ∈R T ; A modified separation transformer, namely SepFormer model, is used, which has a Transformer-based architecture without recurrent neural network RNN, including encoder, decoder and separator. The encoder and separator modules are modified, including the configuration of convolutional layers and activation functions, and the convolution kernel and step size are reduced during the convolution process, as follows: (1) Encoder The encoder adopts a bottleneck design, replacing the one-dimensional convolution, through a double convolution layer, initially increasing and then reducing the number of kernels to accurately extract the coding features of the signal; the encoder receives the time domain mixed signal x and generates the learned two-dimensional feature representation h∈R F×T′ , where F is the feature dimension, T′ is the new length of the signal, and a convolutional layer is used: h = conv1d(ReLU(conv1d(x))), where conv1d() represents a one-dimensional convolution and ReLU() represents a nonlinear activation function; (2) Separator The separator receives the encoded 2D feature representation h and estimates a mask m1,...,m for each source in the mixture. N , that is: m k =separator(h), k=1,...,N; where separator() represents the separator, and N represents the number of signal sources; In the process of estimating the mask, the separator first processes the 2D feature representation h into overlapping 3D feature blocks h′ and inputs h′ into the separation transform block: Where F represents the feature dimension, C is the length of each block, and T c is the number of blocks; The separation transformer block is the core of the separator. It uses a dual-scale approach to simulate short-term and long-term dependencies. It consists of two Transformers: the inner loop Transformer, i.e., IntraT, and the outer loop Transformer, i.e., InterT. IntraT simulates short-term dependencies within each block, and InterT simulates long-term dependencies across blocks after permuting the dimensions. The overall transformation of the separated transform block is defined as: Among them, f intra and f inter They represent IntraT and InterT transformations, respectively, P represents the permutation operation, h′ represents the three-dimensional feature block, and h″ represents the feature after separation transformation; (3) Decoder The decoder uses the transposed convolutional layers and the masks predicted by the mask network to reconstruct the separated signals in the time domain: in, represents the separated k-th source, conv1d-transpose() represents one-dimensional inverse convolution, m k Represents the mask of the k-th source.

4. The small sample single channel source separation anti-interference method based on multi-task learning according to claim 3 is characterized in that: The joint training objective function for creating multi-task learning described in step A3 is as follows: During the training phase, the parameters of the encoder, decoder, and separator are shared, and a unified objective function is used to minimize the combined loss of all tasks, namely: Where J(θ) represents the total loss of the training task, θ represents the parameter set shared by all tasks, and α j is the loss L assigned to the jth task j If all tasks are equally important, then α j =1 / M; The parameters of the separation model are b Optimize across multiple tasks in an alternating manner, namely: Among them, β represents the learning rate, Indicates gradient descent along θ; The loss L for each task i The scale-invariant signal-to-noise ratio SI-SNR of the sentence-level permutation-invariant uPIT loss is used as the loss function, and the scale-invariant signal-to-noise ratio SI-SNR is expressed as: Among them, s k and represents the source signal and the estimated signal, <·,·> represents the dot product of two vectors, ||·||2 represents the two-norm, s target represents the projection component of the estimated signal in the direction of the source signal, e noise Represents the vertical component of the estimated signal in the direction of the source signal.

5. The small sample single channel source separation anti-interference method based on multi-task learning according to claim 1 is characterized in that: Based on the joint training objective function described in step A4, the signal separation network of step A2 is trained using the signal data set obtained in step A1 to obtain a trained pre-trained signal separation network, as follows: The data of the multi-task training set obtained in step A1 is packaged, the sequence is combined into a batch, and sent to the signal separation network in step A2 for training. The network parameters are set, and the Adam optimizer is used with an initial learning rate of β = 0.0125. A total of 100 rounds of training are performed; After training, save the optimal parameter model.

6. The small sample single channel source separation anti-interference method based on multi-task learning according to claim 1 is characterized in that: The creation of the fine-tuning data set described in step B1 includes the communication signal and the small sample target interference signal and the mixed signal of the two, as follows: The target communication signal and the small sample target interference signal are obtained by either of two methods: The first method: computer software is used to generate communication signals and small sample interference signals. The communication signal modulation method is digital modulation or analog modulation. The interference signals include digital modulation interference signals, analog modulation interference signals, noise frequency modulation interference signals, noise amplitude modulation interference signals, single-tone interference signals, multi-tone interference signals, comb spectrum interference signals, broadband noise interference, and swept frequency interference. Based on the generated target communication signal and interference signal, the target communication signal and the target interference signal are randomly mixed to form an aliased signal, and noise is added to the aliased signal to generate an aliased communication signal with a signal-to-noise ratio (SNR) of -5dB to 20dB; The second method is to sample the actual communication equipment and the interference equipment to obtain the data sets of the target communication signal, the target interference signal and the aliased signal; The acquired target communication signal is the same as the communication signal in step A1, the interference signal is of a different type from the interference signal in step A1, and the number of small samples is 0-10.

7. The small sample single channel source separation anti-interference method based on multi-task learning according to claim 1 is characterized in that: The objective function for creating the fine-tuning phase described in step B2 is as follows: SI-SNR with uPIT loss is adopted as the fine-tuning loss function.

8. The small sample single channel source separation anti-interference method based on multi-task learning according to claim 1 is characterized in that: Based on the objective function of the fine-tuning stage described in step B3, the pre-trained signal separation network obtained in step A4 is fine-tuned using the fine-tuning data set obtained in step B1 to obtain a fine-tuned signal separation network, which is specifically as follows: Use a small sample data set to fine-tune the pre-trained model. The fine-tuning rounds are set to 20 rounds. During the fine-tuning process, choose to freeze the encoder, decoder, or separator of the pre-trained model. After fine-tuning, save the optimal model.

9. A small sample single channel source separation anti-interference device based on multi-task learning, characterized in that: The device is used to implement the small sample single-channel source separation anti-interference method based on multi-task learning according to any one of claims 1 to 8, and the device includes a multi-task training module, a small sample fine-tuning module and an application module, wherein: The multi-task training module includes a data set creation unit, a signal separation network construction unit, an objective function creation unit, and a network training unit, wherein: A data set creation unit, which creates a multi-task training data set, including communication signals and various interference signals; Signal separation network construction unit, building encoder, separator and decoder modules; Joint training objective function creation unit, creating a customized multi-task learning training objective function; The network training unit uses the signal data set obtained by the data set creation unit to train the network obtained by the signal separation network construction unit based on the joint training objective function, and obtains a trained pre-trained separation network; The small sample fine-tuning module includes a small sample data set creation unit, a fine-tuning objective function creation unit and a fine-tuning network unit, wherein: A small sample data set creation unit, which creates a fine-tuning data set, including a communication signal and a small sample target interference signal and a mixed signal of the two; A fine-tuning objective function creation unit is used to create an objective function in the fine-tuning stage; Fine-tune the network unit, based on the objective function of the fine-tuning stage, and use the fine-tuning dataset to fine-tune the trained pre-trained separation model to obtain a fine-tuned separation network; The application module includes a signal acquisition unit, a signal separation unit, and a demodulation unit, wherein: A signal acquisition unit, which acquires the time-frequency aliased signal to be separated through a single antenna; The signal separation unit uses a fine-tuned separation network to separate the target communication signal and the interference signal from the aliased signal, thereby completing the separation of the aliased signal; The demodulation unit sends the separated communication signal obtained by the signal separation unit to the demodulator for demodulation to restore the bit data of the signal.

10. A mobile terminal comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the small sample single-channel source separation and anti-interference method based on multi-task learning as described in any one of claims 1 to 8 is implemented.

Images