Signal estimation model training method, signal estimation method and device

Abstract

The invention provides a training method of a signal estimation model and a signal estimation method and device.The method comprises the steps that a signal obtained by sampling a signal received by an antenna through a radio frequency chip in terminal equipment serves as a signal sample, the randomness and authenticity of sample data are improved, the signal estimation model is trained through a training sample, and the signal estimation accuracy is improved. According to the method, the deep learning model has the signal estimation capability, training of the signal estimation model through the training sample belongs to data driving, does not depend on specific radio frequency damage classification and modeling, and is not modified along with change of radio frequency hardware, so that the limitation of scene use is reduced, and better universality is achieved.

Description

Technical Field

[0001] The present application relates to the field of signal processing technology, and in particular to a signal estimation model training method, a signal estimation method and a device. Background Art

[0002] In the field of communications, signal distortion occurs during air interface signal transmission. Distortion includes the addition of noise or distortion. In order to ensure the sensitivity and anti-interference capability of the baseband demodulation capability, the distorted signal is estimated through the digital front end in the terminal device, that is, the distorted signal is corrected to obtain an estimated signal.

[0003] In the related art, signal estimation is implemented by a digital front end in a radio frequency chip, and the signal estimation algorithm implemented by the digital front end is driven by mathematical modeling. The algorithm has limitations and poor versatility. Summary of the invention

[0004] The present application aims to solve one of the technical problems in the related art at least to some extent.

[0005] To this end, the present application proposes a signal estimation model training method, a signal estimation method and a device, which perform signal estimation based on a data-driven neural network model, reduce the limitations of the algorithm and improve the versatility.

[0006] In one aspect, an embodiment of the present application provides a method for training a signal estimation model, comprising:

[0007] Acquire a signal sample; wherein the signal sample is obtained by sampling the signal received by the antenna by the radio frequency chip;

[0008] Performing signal estimation on the signal sample by using a signal estimation model set in the radio frequency chip to obtain an estimated signal;

[0009] The signal estimation model is trained according to the estimation signal and the true value signal corresponding to the signal sample to obtain the trained signal estimation model.

[0010] Another aspect of the present application provides a signal estimation method, including:

[0011] Get the signal to be processed;

[0012] The signal estimation model obtained through training is used to perform signal estimation on the signal to obtain a target signal; wherein the signal estimation model is obtained through training using the method described in the above aspect.

[0013] Another aspect of the present application provides a signal estimation model training device, including:

[0014] An acquisition module is used to acquire a signal sample; wherein the signal sample is obtained by sampling the signal received by the antenna by the radio frequency chip;

[0015] A processing module, configured to perform signal estimation on the signal sample by using a signal estimation model set in the radio frequency chip to obtain an estimated signal;

[0016] The training module is used to train the signal estimation model according to the estimated signal and the true value signal corresponding to the signal sample to obtain the trained signal estimation model.

[0017] Another aspect of the present application provides a signal estimation device, including:

[0018] An acquisition module, used for acquiring a signal to be processed;

[0019] An estimation module is used to use a trained signal estimation model to perform signal estimation on the signal to obtain a target signal; wherein the signal estimation model is trained using the training device for the signal estimation model described in the other aspect above.

[0020] Another aspect of the present application provides an electronic device, 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 method described in the first aspect and the method described in the other aspect are implemented.

[0021] Another aspect of the present application provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method described in the first aspect and the method described in the other aspect are implemented.

[0022] Another aspect of the present application provides a chip, the chip comprising a processing circuit, the processing circuit being configured to execute the method described in the aforementioned first aspect and the method described in the aforementioned other aspect.

[0023] Another aspect of the present application provides a computer program product on which a computer program is stored. When the program is executed by a processor, the method described in the first aspect and the method described in the other aspect are implemented.

[0024] The signal estimation model training method, signal estimation method and device proposed in the present application use the signal obtained by sampling the signal received by the antenna by the RF chip in the terminal device as the signal sample, improve the randomness and authenticity of the sample data, and train the signal estimation model through the training samples, so that the deep learning model has the ability to estimate the signal. The training of the signal estimation model through the training samples is data-driven, does not rely on the classification and modeling of specific RF damage, and is not modified with the change of RF hardware. Therefore, it reduces the limitations of scene use and has better versatility.

[0025] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0027] Figure 1 A schematic diagram of the structure of a receiver in the related technology provided in an embodiment of the present application;

[0028] Figure 2 A flowchart of a method for training a signal estimation model provided in an embodiment of the present application;

[0029] Figure 3 A flowchart of another method for training a signal estimation model provided in an embodiment of the present application;

[0030] Figure 4 A schematic diagram of a digital radio frequency front-end architecture provided in an embodiment of the present application;

[0031] Figure 5 A timing relationship diagram of input and output of a signal estimation model provided in an embodiment of the present application;

[0032] Figure 6 A schematic diagram of the structure of a multi-head attention layer provided in an embodiment of the present application;

[0033] Figure 7 One of the training schematic diagrams of a signal estimation model provided for the example of this application;

[0034] Figure 8 A schematic diagram of a flow chart of a signal estimation method provided in an embodiment of the present application;

[0035] Fig. 9 A schematic diagram of a flow chart of another signal estimation method provided in an embodiment of the present application;

[0036] Fig.10A second training diagram of a signal estimation model provided in an embodiment of the present application;

[0037] Fig.11 A schematic diagram of the structure of a training device for a signal estimation model provided in an embodiment of the present application;

[0038] Fig.12 A schematic diagram of the structure of a signal estimation device provided in an embodiment of the present application;

[0039] Fig.13 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0040] Embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0041] The following describes the signal estimation model training method, signal estimation method and device of the embodiments of the present application with reference to the accompanying drawings.

[0042] Figure 1 A schematic diagram of the structure of a receiver in the related art provided in the embodiment of the present application, such as Figure 1 As shown, the air interface signal is coupled to the RF transmission path through the antenna (Antenna) 101, and first passes through the RF front end (RF Front End) module 102. The RF front end module 102 usually includes devices such as a duplexer, an antenna switch, and a low noise amplifier (LNA). The RF front end module 102 can be replaced according to different RF frequency band requirements. After the RF module 102 processes the air interface signal, the output signal is sent to the RF processor (RF Processor) 103, and the analog front end (Analog Front End) 104 mixes the signal to the baseband and performs filtering and analog-to-digital conversion. Since the baseband processing requirements do not change with the RF frequency band, the same analog front end 104 can be adapted to different RF front end modules 102 to facilitate modular processing of different RF frequency bands. After being converted into a digital signal by the analog front end 104, the digital front end (DFE) 105 estimates the digital signal to complete tasks such as correction of radio frequency damage, and finally sends the estimated signal to the baseband processor (Base-Band Processor) 106 for demodulation.

[0043] Among them, the digital front end (Digital Front End) 105 has the following problems in estimating the signal:

[0044] First, the architecture of the digital front end is not universal enough. The digital front end needs to be built on a specific signal modeling basis and is bound to the design architecture. The digital front end is usually implemented by an application-specific integrated circuit (ASIC), and the digital module in the chip cannot be changed. When there are changes such as RF front end or architecture upgrades, it will have to be redesigned and re-taped, which requires a high cost.

[0045] The second point is the limitation of the algorithm model. The design of the RF non-ideal algorithm cannot cope with RF damage beyond the expected modeling.

[0046] To solve the above problems, in the embodiments of the present application, a signal estimation algorithm is implemented through a deep learning signal estimation model, and the signal obtained by sampling the signal received by the antenna by the RF chip in the terminal device is used as a signal sample to improve the randomness and authenticity of the sample data. The signal estimation model is trained through training samples, so that the deep learning model has the ability to estimate signals. Training the signal estimation model through training samples is data-driven, does not rely on the classification and modeling of specific RF damage, and is not modified with changes in RF hardware. Therefore, it reduces the limitations of scenario use and has better versatility.

[0047] Figure 2 A flowchart of a method for training a signal estimation model provided in an embodiment of the present application.

[0048] As an implementation method, the training method of the signal estimation model of the embodiment of the present application can be configured in a training device of the signal estimation model, and the training device of the signal estimation model can be applied to any electronic device or chip so that the electronic device can perform the training function of the signal estimation model. Among them, the chip is, for example, a microprocessor (General Parallel Unit, GPU) with a highly parallel computing architecture.

[0049] The electronic device may be any device with computing capability, for example, a computer, and the computer may be a server, such as a local server.

[0050] like Figure 2 As shown, the method may include the following steps:

[0051] Step 211, obtaining signal samples.

[0052] In an embodiment of the present application, the signal sample includes an air interface signal, which may be a wireless signal in the field of wireless communication, such as a wireless signal in the field of 5G or 4G communication, or a radar signal in the radar field, or a wireless signal in the field of Internet of Things.

[0053] In an embodiment of the present application, the signal sample can be obtained by sampling the signal received by the antenna by the RF chip in the terminal device. As an implementation method, the terminal device receives the air interface signal through the antenna, and sends it to the RF chip after processing by the RF front end. The RF chip samples the air interface signal processed by the RF front end to obtain the signal sample, and the signal obtained by sampling the air interface signal received in the actual scene is used as the signal sample, wherein the RF damage in the signal sample has stronger authenticity, randomness and complexity than the artificially added RF damage, thereby improving the offline training effect of the model, and the training of the signal estimation model driven by the training data makes the signal estimation model process any type of air interface signal and any type of RF damage through learning, and is not affected by the RF hardware, such as the change of the analog front end and the RF module. Even in the case of changes in the analog front end and the RF module, resampling the data for training can adapt. At the same time, through the deep learning characteristics of the signal estimation model, the nonlinear distortion that cannot be represented by mathematical modeling in the related technology can be learned, which improves the signal estimation ability of the model and improves the versatility of the signal estimation model.

[0054] In one implementation of the embodiment of the present application, the signal sample is a signal sequence, and the signal sequence includes air interface signals sampled in multiple time units, thereby achieving synchronous processing of multiple signals and improving processing efficiency.

[0055] Step 212: perform signal estimation on the signal sample by using a signal estimation model set in the radio frequency chip to obtain an estimated signal.

[0056] In the embodiment of the present application, the signal estimation model integrates DFE, which is used to estimate the received signal. Since the signal received by the antenna has problems such as distortion and noise interference, in order to ensure the sensitivity and anti-interference capability of the subsequent baseband chip demodulation capability, it is necessary to complete the signal estimation process of the signal through the signal estimation model, wherein signal estimation refers to estimating various damages of the signal, calculating compensation information based on various losses, and compensating the received signal according to the compensation information to obtain an estimated signal.

[0057] In an embodiment of the present application, a signal estimation model is provided in a radio frequency chip of an electronic device. Since there are hardware deviations between different batches of radio frequency chips, it is necessary to train the signal estimation model provided in the radio frequency chip in the electronic device. By training the signal estimation model, the trained signal estimation model can realize mapping between the original signal and the estimated signal, so as to remove distortion information such as radio frequency damage in the signal and obtain an estimated signal. The estimated signal is a signal without distortion after distortion information such as radio frequency damage in the signal sample is removed. Since the neural network model does not depend on the modeling form of specific radio frequency damage, nor on a specific radio frequency front-end architecture, the estimation and compensation of radio frequency damage is realized through data acquisition, thereby improving the versatility of signal processing.

[0058] Step 213: Train the signal estimation model according to the estimated signal and the true value signal corresponding to the signal sample to obtain a trained signal estimation model.

[0059] In the embodiment of the present application, the difference between the estimated signal and the true value signal corresponding to the signal sample is determined, and the loss function is determined according to the difference. As an implementation method, the loss function may be a loss function determined based on a two-norm (L2 Loss), or a loss function determined based on a one-norm (L1 Loss);

[0060] Among them, the loss function expressed by a norm is as follows:

[0061]

[0062] Among them, the loss function expressed by the two norm is as follows:

[0063]

[0064] Among them, s i is a true value signal, To estimate the signal, σ is the standard deviation of the dataset of training samples, which is determined when constructing the dataset of training samples.

[0065] The training method of the signal estimation model of the embodiment of the present application obtains signal samples, estimates the signal samples through the signal estimation model to obtain an estimated signal, trains the signal estimation model according to the estimated signal and the true signal carried in the signal sample to obtain the trained signal estimation model, implements the signal estimation function through the deep learning signal estimation model, and uses the RF chip in the terminal device to sample the signal received by the antenna to obtain the signal as the signal sample, thereby improving the randomness and authenticity of the sample data, and trains the signal estimation model through the training samples so that the deep learning model has the ability to estimate the signal. The training of the signal estimation model through the training samples is data-driven, does not rely on the classification and modeling of specific RF damage, and is not modified with the change of RF hardware. Therefore, it reduces the limitations of scenario use and has better versatility.

[0066] Based on the above embodiments, Figure 3 A flow chart of another method for training a signal estimation model provided in an embodiment of the present application is shown in FIG. Figure 3 As shown, the method comprises the following steps:

[0067] Step 301, obtaining signal samples.

[0068] In the implementation of this application, the signal sample is used as a signal sequence for illustration. As an example, the model training method of the embodiment of this application can be applied to the scenario of factory calibration of terminal equipment to realize calibration of signal estimation models for different terminal equipment and improve the accuracy of the signal estimation models configured in each terminal equipment.

[0069] In the embodiment of the present application, the digital RF front-end architecture includes an RF front-end module 201, an analog front-end 202 and a signal estimation model 203. The signal estimation model 203 obtained through training realizes the signal estimation function of the digital front-end DFE, and the signal estimation model 203 is deployed in the RF chip. The air interface signal passes through the RF front-end module 201 and the analog front-end 202 and is sent to the signal estimation model 203, wherein the RF front-end module 201 completes the signal filter and amplification (such as low noise amplifier - LNA), and the analog front-end 202 completes the conversion of analog signals to digital signals through down sampling, low pass filtering LPF, and analog-to-digital conversion ADC to adapt to the requirements of the baseband signal. As an example, Figure 4 As shown, the implementation methods of the RF front-end module 201 and the analog front-end 202 are not unique and are not limited in this embodiment.

[0070] As an example, after analog-to-digital conversion, the air interface signal is converted into an in-phase and quadrature signal, i.e., an I / Q signal. If the signal sample includes N signals, i.e., the antenna receives N signals in different receiving time units, the N signals are filled into a cache with a depth of N, so that the signal estimation model processes N I / Q signals at a time, that is, the dimension of the signal sample is N×2, i.e., the signal sample is {x i,j} N×2 , where i and j are used to indicate the data of the I dimension and Q dimension in the signal sample. As an example, Figure 4 As shown, the signal estimation model 203 includes a coding network 2031 , a sampling network 2032 , an encoder 2033 and an output network 2034 .

[0071] Step 302: Use a coding network to perform position coding on the signal sequence to obtain a first coding feature.

[0072] In the implementation of the present application, since the signal sample is a signal sequence, the signal sequence includes signals received by each receiving time unit. In order to be able to identify the receiving order of each signal in the output estimated signal after the signal estimation model is processed, position information is added to each signal in the signal sequence, that is, position coding is implemented. As an implementation method, the receiving time unit of each signal included in the signal sequence is obtained, and the position information of each signal in the signal sequence is determined according to the receiving time unit of each signal, wherein the position information is used to indicate the receiving order of each signal in the signal sequence, and the corresponding position information is added to each signal in the signal sequence to obtain the first coding feature.

[0073] As an implementation method, the location information can be added by the following formula:

[0074]

[0075] Among them, x' i,j is a signal in the signal sequence, p i,j is the position information corresponding to a signal in the signal sequence, x i,j It is a signal with the position information added, and L is a set constant, which can be set according to the scene requirements and is not limited in this embodiment.

[0076] Step 303: Use a sampling network to sample the first coding feature to obtain a second coding feature.

[0077] Among them, the second coding feature is a feature that matches the processing dimension of the radio frequency chip.

[0078] In the embodiment of the present application, since the dimension of the signal processing of the RF chip and the dimension of the input data may be different, it is necessary to perform sampling processing according to the processing dimension of the RF chip, and the sampling can be upsampling or downsampling.

[0079] As an example, sampling is usually downsampling. The sampling network is composed of a 1-2 layer fully connected network, which is not limited in this embodiment. For example, the signal estimation model works at a D1 times downsampling rate, and the output rate of the estimation result works at a D1D2 times downsampling rate. Then the dimensionality of the signal sequence in the signal estimation model changes as follows: Figure 4 As shown in the mark: the input of the encoding network 2031 is a signal sequence of N signals, which is sent to the sampling network 2032 with a dimension of N×2 after encoding; the output dimension of the sampling network 2032 is N / D1×2, and the dimension is maintained unchanged in the encoder 2033 until it is sent to the output layer 2034. The final dimension after processing by the output layer 2034 is N / (D1D2)×2.

[0080] As an example, Figure 5 A timing diagram of the input and output of a signal estimation model provided in an embodiment of the present application, such as Figure 5 As shown, at the time t0 of the digital clock, the signal sequence S1 of N signals is processed, and the estimated signal sequence S2 of M signals is output (where M=N / (D1D2)). Then, at the time tM' of the digital clock, the signal sequence S3 of N signals is processed, and the estimated signal sequence S4 of M signals is output, where M=N / (D1D2). Under the same digital clock Fs, N points are input and M points are output. The sampling rate of the actual corresponding input signal samples is Fs*N, and the sampling rate of the output estimated signal is Fs*M, thereby realizing the processing of signals of different rates.

[0081] Step 304: Use an encoder to perform signal estimation based on the second coding feature to obtain a target signal feature.

[0082] In one implementation of the embodiment of the present application, the encoder includes a multi-head attention layer and a residual connection layer, and the multi-head attention layer in the encoder is used to perform attention encoding on the second coding feature to obtain a third coding feature output by the multi-head attention layer, wherein the third coding feature carries the estimated signal information, and then, the residual connection layer in the encoder is used to fuse and normalize the third coding feature and the second coding feature to obtain the target signal feature, wherein the fusion here can be splicing. Among them, the second coding feature and the third coding feature are fused because the gradient of the second coding feature is relatively stable, and it will not be impossible to perform back propagation when the subsequent gradient changes to zero, so as to ensure the smooth progress of the training process. In the embodiment of the present application, the training samples of the signal sequence containing multiple signals can be processed in parallel, which improves the processing effect compared to the related art that processes a single data each time. Due to the increase in sequence length, the multi-head attention layer is used to process the longer signal sequence, and the multi-head attention layer introduces the front and back dependencies of the signal sequence, which can flexibly process the longer signal sequence, and it is easier to implement parallel training on the time signal to improve the training effect.

[0083] As an example, Figure 6 A schematic diagram of the structure of a multi-head attention layer provided in an embodiment of the present application is shown in FIG. Figure 6 As shown, the N / D1×2 dimensional second coded data X in 401 Input multi-head attention layer, through W Q , W K , W V Operators (Query 402, Key 403 and Value 404) are used to encode the information between different positions in the second encoder feature, where the dimension of the operator is 2×Da. The output result y of a single attention head is obtained through the dot product Dot Product and norm operation in 405. h The operation relationship is:

[0084]

[0085] The softmax operation corresponding to 406 can be implemented in a variety of approximate ways, which are not limited in the embodiments of the present application. The classic softmax operation is as follows:

[0086]

[0087] In this example, there are H attention heads, where H is a natural number greater than or equal to 2. All attention results are concatenated to get the final output 407 in the form of:

[0088] Y=[y1,…,y H ];

[0089] The dimension Da and the dimension H satisfy the relationship H×Da=2, ensuring that the dimension of the final concatenated Y is the same as the input X. in The concatenated data is then transformed by a weight matrix W of dimension N / D1×N / D1 to obtain the final output third encoding feature X out , the formula is as follows:

[0090] X out =WY;

[0091] Signal X after multi-head attention processing out It is sent to the residual connection layer to complete the residual connection, that is, the third encoding feature and the second encoding feature are fused, and the formula for the fusion result X is as follows:

[0092] X=Norm(X out +X in );

[0093] Among them, the residual connection layer can also perform a normalization Norm operation on the fusion result X obtained by the residual connection, and obtain the target signal characteristics through the normalization operation. Among them, the classic layer normalization is implemented using the following formula:

[0094]

[0095]

[0096] Where a and b are the weights and biases of the signal estimation model, μ and σ 2 are the expectation and variance respectively, and ε is a small fixed constant set to prevent the denominator from being zero.

[0097] Normalization methods include: root mean square normalization, normalization based on NTK parameterization, etc. The normalization method is not limited in the embodiments of the present application.

[0098] In another implementation of the embodiment of the present application, the encoder includes a multi-head attention layer, a residual connection layer, a feedforward layer and a post-stage residual connection. Similarly, after completing the multi-head attention layer processing and the residual connection layer fusion, the output of the encoder can be obtained through the feedforward layer and the post-stage residual connection. Among them, the structure of the post-stage residual connection layer is exactly the same as the structure of the residual connection layer. The feedforward layer can be composed of a fully connected network or implemented by other forms of networks, as long as the output dimension and the input dimension remain consistent.

[0099] Among them, Figure 4As shown, encoder 2033 can be one or more, such as K, where K is a natural number greater than or equal to 1. As an implementation method, when there are multiple encoders, multiple encoders are processed in parallel to achieve encoding from different interference or distortion angles to obtain diverse encoding results, and the target signal features output by multiple encoders are fused to obtain the target signal features of the final input-output network to improve the accuracy of the encoding results.

[0100] Step 305: Use the output network to obtain an estimated signal according to the target signal characteristics.

[0101] In an embodiment of the present application, the output network extracts an estimated signal based on the characteristics of the target signal. As an implementation method, in order to adapt the baseband signal to the data processing requirements, the estimated signal can be sampled and interface-adapted, wherein the sampling is downsampling to achieve data matching to facilitate subsequent baseband chip processing.

[0102] Step 306: Train the signal estimation model according to the estimated signal and the true value signal carried in the signal sample to obtain a trained signal estimation model.

[0103] In one implementation method of the embodiment of the present application, the training process of the signal estimation model can be regarded as a classification task. According to the bit width of the estimated signal, the first probability distribution corresponding to the estimated signal is determined, and the second probability distribution corresponding to the estimated signal is determined. According to the difference between the first probability distribution and the second probability distribution, the loss function is determined. The signal estimation model is trained according to the loss function to obtain a trained signal estimation model.

[0104] As an example, Figure 7 One of the training diagrams of a signal estimation model provided for the example of this application, such as Figure 7 As shown, the signal source 501 inputs the true value signal s(n) of the signal sample into the RF chip 502, and the DFE samples the input data x(n) and groups it with s(n) and sends it to the host computer 502 (usually a PC containing a General Parallel Unit, GPU). The host computer first initializes the parameters Φ510 of the signal estimation model, and then standardizes {x(n), s(n)} in batches to ensure the stability of the values ​​and enhance the convergence speed and generalization ability of the model. Subsequently, forward propagation and backward propagation iterations are implemented batch by batch in the host computer until the loss function converges. Finally, the trained network parameters are written into the chip registers.

[0105] The bit width of the output estimated signal is set to 12 bits. Then, for the estimated signal corresponding to the nth signal sample, The value needs to be in 2 12 Therefore, the classic cross entropy can be used as the loss function, as follows:

[0106]

[0107] in, is the cross entropy loss function, Q i is the characteristic function of the fixed-point data category of the data, and s(n) is the nth signal sample.

[0108] Then, when the loss function is determined, the data output by the intermediate network of the forward propagation is read, including the output layer gradient 507, the encoding layer gradient 508 and the input layer gradient 509, and the back propagation gradient J is calculated layer by layer. Loss , update the network weight Φ510 according to the back propagation result. The training process can be performed multiple times based on different signal samples until the loss function is less than the set threshold, or the number of training times reaches the set value, and the model training is completed.

[0109] In the training method of the signal estimation model of the embodiment of the present application, in the existing digital front-end architecture, most designs rely on mathematical modeling of RF damage, and different RF modules and analog front-end designs will lead to different RF damage modeling methods, which in turn affect the design of the digital front-end architecture. Therefore, these digital front-end designs are not universal. The universal digital front-end architecture based on the multi-head attention mechanism proposed in this application is a signal estimation model obtained based on data training. The signal estimation model does not depend on the modeling form of specific RF damage, nor does it depend on a specific RF front-end architecture. It can estimate and compensate for RF damage only through data acquisition. It is also suitable for variable-rate time series processing and has better versatility. At the same time, for other neural network models, the training data comes from artificially added RF damage, while the training data of this application comes from hardware sampling, which reduces limitations, can meet the needs of different situations, and improves versatility.

[0110] Based on the above embodiments, in one implementation of the embodiments of the present application, there are multiple encoders, and each encoder is used to perform signal estimation according to the second coding feature to obtain the first signal feature corresponding to each encoder, and multiple first signal features are fused to obtain the target signal feature. Multiple encoders are used to perform signal estimation according to the second coding feature, and signal features of multiple different angles or dimensions can be obtained. After fusion, these features can more comprehensively reflect the true situation of the signal, thereby improving the accuracy of signal estimation. Especially when processing complex RF signals, a single encoder may not be able to fully capture all the features of the signal, and the combined use of multiple encoders can more effectively deal with this situation.

[0111] Based on the above embodiments, Figure 8 A flow chart of a signal estimation method provided in an embodiment of the present application. The executor of the embodiment of the present application is a terminal device, such as a mobile phone, a pad, a smart wearable device, etc., which is not limited in this embodiment.

[0112] like Figure 8 As shown, the method comprises the following steps:

[0113] Step 801: Obtain a signal to be processed.

[0114] In one scenario of an embodiment of the present application, the signal to be processed is an air interface signal obtained by sampling the RF chip of the terminal device, wherein the relevant explanations and instructions in the aforementioned embodiments are also applicable to this embodiment, and the principles are the same and will not be repeated here.

[0115] Step 802: Use the trained signal estimation model to perform signal estimation on the signal to obtain a target signal.

[0116] The signal estimation model is obtained by training using the training method of the signal estimation model in any of the aforementioned embodiments. The target signal, relative to the signal to be processed, is a signal with radio frequency damage corrected. The training process can refer to the relevant explanations in the aforementioned embodiments, and the principle is the same, so it will not be repeated here.

[0117] In the signal estimation method of the embodiment of the present application, the signal estimation function of the digital front end DFE is realized through a deep learning signal estimation model, and the signal obtained by sampling the signal received by the antenna using the RF chip in the terminal device is used as a signal sample to improve the randomness and authenticity of the sample data. The signal estimation model is trained through training samples so that the deep learning model has the ability to estimate signals. The training of the signal estimation model through training samples is data-driven, does not rely on the classification and modeling of specific RF damage, and is not modified with the change of RF hardware. Therefore, it reduces the limitations of scene use and has better versatility. Furthermore, the signal estimation model obtained by training is used to estimate the signal to be processed, so that the signal estimation result can be obtained quickly and accurately, which improves the versatility of scene use.

[0118] Based on the above embodiments, Fig. 9 A flow chart of another signal estimation method provided in an embodiment of the present application specifically illustrates how the signal estimation model is updated online during actual use, such as Fig. 9 As shown, for the online updating process of the signal estimation model, the method includes:

[0119] Step 901: Acquire signal samples, input the signal samples into a baseband processor for signal extraction, and obtain a predicted pilot signal.

[0120] The signal sample is obtained by sampling the signal received by the antenna through the radio frequency chip during the use of the terminal device. For details, please refer to the description of the signal sample in the above embodiment. The principle is similar and will not be repeated here.

[0121] In the embodiment of the present application, the signal sample is input into the baseband processor, and the baseband processor extracts the target signal to obtain the predicted pilot signal.

[0122] Step 902: Determine a loss function based on the predicted pilot signal and the reference pilot signal stored in the terminal device.

[0123] The reference pilot signal may be an ideal pilot signal pre-stored in the terminal device.

[0124] Step 903: Update the signal estimation model according to the loss function to obtain an updated signal estimation model.

[0125] In an embodiment of the present application, a loss function is determined based on the difference between the predicted pilot signal and the reference pilot signal stored in the RF chip, data output by each network in the signal estimation model is obtained, a back propagation gradient is calculated based on the loss function and the data output by each network, and parameters of the signal estimation model are updated based on the back propagation gradient to adapt to signal estimation requirements in actual scenarios, thereby achieving online updating of the signal estimation model and improving the accuracy of the model.

[0126] As an example, Fig.10 A second training diagram of a signal estimation model provided in an embodiment of the present application is shown in FIG. Fig.10 As shown, after the receiver RFIC (Radio Frequency Integrated Chip, RFIC) inputs the signal InputSignal601, the output network of the signal estimation model outputs the target signal 604. This process is equivalent to the forward propagation operation. The target signal is extracted through the baseband processor 602 (Base-band Integrated Chip, BBIC) to obtain the pilot signal. And obtain the expected ideal pilot signal s 605, where the ideal pilot signal matches the communication protocol. s and the intermediate variable Y output by each network (coding network, sampling network, encoder) of the signal estimation model are sent to the processing module 603 in the terminal device. For example, the processing module 603 is a neural network processing unit (NPU), which is used to calculate the loss function and perform back propagation. After completion, the updated parameter Φ is configured into the signal estimation model to realize online real-time update of the signal estimation model.

[0127] In an embodiment of the present application, the signal estimation model is updated online. In one scenario, when the terminal device detects that the difference between the current ambient temperature and the temperature during the production line calibration is greater than the set temperature threshold, it is considered that the RF characteristics will change, triggering the online update of the signal estimation model; in a second scenario, when the baseband signal of the terminal device determines that the bit error rate of the demodulation result of the signal is greater than the set threshold, the online update of the signal estimation model is triggered; in the third scenario, after the terminal device has been used for a set period of time, the online update of the signal estimation model is triggered based on hardware aging, so as to realize timely update of the signal estimation model and ensure the effect of signal estimation.

[0128] In order to implement the above embodiment, the embodiment of the present application also proposes a training device for a signal estimation model.

[0129] Fig.11 A schematic diagram of the structure of a training device for a signal estimation model provided in an embodiment of the present application.

[0130] like Fig.11 As shown, the device may include:

[0131] The acquisition module 110 is used to acquire signal samples; wherein the signal samples are obtained by sampling the signal received by the antenna by the radio frequency chip.

[0132] The processing module 111 is used to perform signal estimation on the signal sample by using a signal estimation model set in the radio frequency chip to obtain an estimated signal.

[0133] The training module 112 is used to train the signal estimation model according to the estimated signal and the true value signal corresponding to the signal sample to obtain the trained signal estimation model.

[0134] Further, in an implementation of the embodiment of the present application, the signal sample is a signal sequence, the signal estimation model includes a coding network, a sampling network, an encoder and an output network, and the processing module 111 is specifically used to:

[0135] Using the encoding network to position encode the signal sequence to obtain a first encoding feature;

[0136] The first coding feature is sampled using the sampling network to obtain a second coding feature; wherein the second coding feature is a feature that matches the processing dimension of the radio frequency chip;

[0137] Using the encoder to perform signal estimation according to the second coding feature to obtain a target signal feature;

[0138] The output network is used to obtain the estimated signal according to the target signal characteristics.

[0139] In one implementation of the embodiment of the present application, the processing module 111 is specifically configured to:

[0140] Performing attention encoding on the second encoded feature using a multi-head attention layer in the encoder to obtain a third encoded feature output by the multi-head attention layer; wherein the third encoded feature carries the estimated signal information;

[0141] The third encoding feature and the second encoding feature are fused and normalized using a residual connection layer in the encoder to obtain the target signal feature.

[0142] In one implementation of the embodiment of the present application, the processing module 111 is specifically configured to:

[0143] Acquire a reception time unit of each signal included in the signal sequence;

[0144] Determine the position information of each signal in the signal sequence according to the receiving time unit of each signal; wherein the position information is used to indicate the receiving order of each signal in the signal sequence;

[0145] Corresponding position information is added to each signal in the signal sequence to obtain the first coding feature.

[0146] In an implementation of the embodiment of the present application, the encoder is multiple, and the processing module 111 is specifically used to:

[0147] Using each of the encoders to perform signal estimation according to the second encoding feature to obtain a first signal feature corresponding to each encoder;

[0148] A plurality of first signal features are fused to obtain the target signal feature.

[0149] In one implementation of the embodiment of the present application, the training module 112 is specifically configured to:

[0150] Determining, according to the bit width of the estimated signal, a first probability distribution corresponding to the estimated signal, and determining a second probability distribution corresponding to the estimated signal;

[0151] determining a loss function based on a difference between the first probability distribution and the second probability distribution;

[0152] The signal estimation model is trained according to the loss function to obtain the trained signal estimation model.

[0153] It should be noted that the above explanation of the method embodiment is also applicable to the device of this embodiment, and will not be repeated here.

[0154] The signal estimation model training device of the embodiment of the present application obtains signal samples, estimates the signal samples through the signal estimation model to obtain an estimated signal, and trains the signal estimation model according to the estimated signal and the true signal carried in the signal sample to obtain the trained signal estimation model. The signal estimation function of the digital front end DFE is realized through the signal estimation model of deep learning, and the signal obtained by sampling the signal received by the antenna by the RF chip in the terminal device is used as the signal sample to improve the randomness and authenticity of the sample data. The signal estimation model is trained through the training samples so that the deep learning model has the ability to estimate the signal. The training of the signal estimation model through the training samples is data-driven, does not rely on the classification and modeling of specific RF damage, and is not modified with the change of RF hardware. Therefore, the limitations of scene use are reduced and it has better versatility.

[0155] In order to implement the above embodiment, the embodiment of the present application further proposes a signal estimation device.

[0156] Fig.12 A schematic diagram of the structure of a signal estimation device provided in an embodiment of the present application.

[0157] like Fig.12 As shown, the device may include:

[0158] The acquisition module 120 is used to acquire the signal to be processed.

[0159] The estimation module 121 is used to use the trained signal estimation model to perform signal estimation on the signal to obtain the target signal; wherein the signal estimation model is trained by using the training device of the signal estimation model in the above-mentioned embodiment. The training method can refer to the relevant explanations in the above-mentioned embodiment, the principle is the same, and it will not be repeated here.

[0160] In an implementation of the embodiment of the present application, the device further includes an update module, the update module being configured to:

[0161] Acquire a signal sample, and input the signal sample into a baseband processor for signal extraction to obtain a predicted pilot signal;

[0162] Determining a loss function based on the predicted pilot signal and a reference pilot signal stored in the terminal device;

[0163] The signal estimation model is updated according to the loss function to obtain an updated signal estimation model.

[0164] It should be noted that the above explanation of the method embodiment is also applicable to the device of this embodiment, and will not be repeated here.

[0165] In the signal estimation device of the embodiment of the present application, the signal estimation function of the digital front end DFE is realized through a deep learning signal estimation model, and the signal obtained by sampling the signal received by the antenna using the RF chip in the terminal device is used as a signal sample to improve the randomness and authenticity of the sample data. The signal estimation model is trained through training samples so that the deep learning model has the ability to estimate signals. The training of the signal estimation model through training samples is data-driven, does not rely on the classification and modeling of specific RF damage, and is not modified with the change of RF hardware. Therefore, the limitations of scene use are reduced and it has better versatility. Furthermore, the signal estimation model obtained by training is used to estimate the signal to be processed, so that the signal estimation result can be obtained quickly and accurately, which improves the versatility of scene use.

[0166] In order to implement the above embodiments, the present application also proposes a non-temporary computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the method described in the above method embodiments is implemented.

[0167] In order to implement the above embodiments, the present application also proposes a computer program product on which a computer program is stored. When the computer program is executed by a processor, the method described in the above method embodiments is implemented.

[0168] In order to implement the above embodiments, the present application also proposes an electronic device, 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 method described in the above method embodiments is implemented.

[0169] Fig.13 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. For example, the electronic device 800 may be a mobile phone, a computer, a digital broadcast terminal, a message transceiver device, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, etc.

[0170] Reference Fig.13 , the electronic device 800 may include one or more of the following components: a processing component 802 , a memory 804 , a power component 806 , a multimedia component 808 , an audio component 810 , an input / output (I / O) interface 812 , a sensor component 814 , and a communication component 816 .

[0171] The processing component 802 generally controls the overall operation of the electronic device 800, such as operations associated with display, phone calls, data communications, camera operations, and recording operations. The processing component 802 may include one or more processors 820 to execute instructions to complete all or part of the steps of the above-mentioned method. In addition, the processing component 802 may include one or more modules to facilitate the interaction between the processing component 802 and other components. For example, the processing component 802 may include a multimedia module to facilitate the interaction between the multimedia component 808 and the processing component 802.

[0172] The memory 804 is configured to store various types of data to support operations on the electronic device 800. Examples of such data include instructions for any application or method operating on the electronic device 800, contact data, phone book data, messages, pictures, videos, etc. The memory 804 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0173] The power component 806 provides power to the various components of the electronic device 800. The power component 806 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power for the electronic device 800.

[0174] The multimedia component 808 includes a screen that provides an output interface between the electronic device 800 and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touch, slide, and gestures on the touch panel. The touch sensor may not only sense the boundaries of the touch or slide action, but also detect the duration and pressure associated with the touch or slide operation. In some embodiments, the multimedia component 808 includes a front camera and / or a rear camera. When the electronic device 800 is in an operating mode, such as a shooting mode or a video mode, the front camera and / or the rear camera may receive external multimedia data. Each front camera and rear camera may be a fixed optical lens system or have a focal length and optical zoom capability.

[0175] The audio component 810 is configured to output and / or input audio signals. For example, the audio component 810 includes a microphone (MIC), and when the electronic device 800 is in an operating mode, such as a call mode, a recording mode, and a voice recognition mode, the microphone is configured to receive an external audio signal. The received audio signal can be further stored in the memory 804 or sent via the communication component 816. In some embodiments, the audio component 810 also includes a speaker for outputting audio signals.

[0176] I / O interface 812 provides an interface between processing component 802 and peripheral interface modules, such as keyboards, click wheels, buttons, etc. These buttons may include but are not limited to: home button, volume button, start button, and lock button.

[0177] The sensor assembly 814 includes one or more sensors for providing various aspects of status assessment for the electronic device 800. For example, the sensor assembly 814 can detect the open / closed state of the electronic device 800, the relative positioning of the components, such as the display and keypad of the electronic device 800, and the sensor assembly 814 can also detect the position change of the electronic device 800 or a component of the electronic device 800, the presence or absence of contact between the user and the electronic device 800, the orientation or acceleration / deceleration of the electronic device 800, and the temperature change of the electronic device 800. The sensor assembly 814 may include a proximity sensor configured to detect the presence of a nearby object without any physical contact. The sensor assembly 814 may also include an optical sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, the sensor assembly 814 may also include an acceleration sensor, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.

[0178] The communication component 816 is configured to facilitate wired or wireless communication between the electronic device 800 and other devices. The electronic device 800 can access a wireless network based on a communication standard, such as WiFi, 4G or 5G, or a combination thereof. In an exemplary embodiment, the communication component 816 receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 816 also includes a near field communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on radio frequency identification (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology and other technologies.

[0179] In an exemplary embodiment, the electronic device 800 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above methods.

[0180] In an exemplary embodiment, a non-transitory computer-readable storage medium including instructions is also provided, such as a memory 804 including instructions, and the instructions can be executed by a processor 820 of an electronic device 800 to perform the above method. For example, the non-transitory computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.

[0181] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0182] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0183] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.

[0184] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute the instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purpose of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways if necessary, and then stored in a computer memory.

[0185] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0186] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0187] In addition, each functional unit in each embodiment of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0188] The storage medium mentioned above may be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the present application. A person of ordinary skill in the art may change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A method for training a signal estimation model, characterized in that: include: Acquire a signal sample; wherein the signal sample is obtained by sampling the signal received by the antenna by the radio frequency chip; Performing signal estimation on the signal sample by using a signal estimation model set in the radio frequency chip to obtain an estimated signal; The signal estimation model is trained according to the estimation signal and the true value signal corresponding to the signal sample to obtain the trained signal estimation model.

2. The method according to claim 1, characterized in that The signal sample is a signal sequence, the signal estimation model includes a coding network, a sampling network, an encoder and an output network, and the signal estimation model set in the radio frequency chip is used to perform signal estimation on the signal sequence to obtain an estimated signal, including: Using the encoding network to position encode the signal sequence to obtain a first encoding feature; The first coding feature is sampled using the sampling network to obtain a second coding feature; wherein the second coding feature is a feature that matches the processing dimension of the radio frequency chip; Using the encoder to perform signal estimation according to the second coding feature to obtain a target signal feature; The output network is used to obtain the estimated signal according to the target signal characteristics.

3. The method according to claim 2, characterized in that The using the encoder to perform signal estimation according to the second coding feature to obtain a target signal feature includes: Performing attention encoding on the second encoded feature using a multi-head attention layer in the encoder to obtain a third encoded feature output by the multi-head attention layer; wherein the third encoded feature carries the estimated signal information; The third encoding feature and the second encoding feature are fused and normalized using a residual connection layer in the encoder to obtain the target signal feature.

4. The method according to claim 2, characterized in that The using the encoding network to position encode the signal sequence to obtain a first encoding feature includes: Acquire a reception time unit of each signal included in the signal sequence; Determine the position information of each signal in the signal sequence according to the receiving time unit of each signal; wherein the position information is used to indicate the receiving order of each signal in the signal sequence; Corresponding position information is added to each signal in the signal sequence to obtain the first coding feature.

5. The method according to claim 2, characterized in that The encoder is multiple, and the encoder is used to perform signal estimation according to the second coding feature to obtain the target signal feature, including: Using each of the encoders to perform signal estimation according to the second encoding feature to obtain a first signal feature corresponding to each encoder; A plurality of first signal features are fused to obtain the target signal feature.

6. The method according to claim 1, characterized in that The step of training a signal estimation model according to the estimation signal and the true value signal corresponding to the signal sample to obtain the trained signal estimation model comprises: Determining, according to the bit width of the estimated signal, a first probability distribution corresponding to the estimated signal, and determining a second probability distribution corresponding to the estimated signal; determining a loss function based on a difference between the first probability distribution and the second probability distribution; The signal estimation model is trained according to the loss function to obtain the trained signal estimation model.

7. A signal estimation method, characterized in that: include: Get the signal to be processed; The signal estimation model obtained through training is used to perform signal estimation on the signal to obtain a target signal; wherein the signal estimation model is obtained through training using the method described in any one of claims 1 to 6.

8. The method according to claim 6, characterized in that The method further comprises: Get signal samples; Inputting the signal sample into a baseband processor for signal extraction to obtain a predicted pilot signal; Determining a loss function based on the predicted pilot signal and a reference pilot signal stored in the terminal device; The signal estimation model is updated according to the loss function to obtain an updated signal estimation model.

9. A training device for a signal estimation model, characterized in that: include: An acquisition module is used to acquire a signal sample; wherein the signal sample is obtained by sampling the signal received by the antenna by the radio frequency chip; A processing module, configured to perform signal estimation on the signal sample by using a signal estimation model set in the radio frequency chip to obtain an estimated signal; The training module is used to train the signal estimation model according to the estimated signal and the true value signal corresponding to the signal sample to obtain the trained signal estimation model.

10. A signal estimation device, characterized in that: include: An acquisition module, used for acquiring a signal to be processed; An estimation module is used to use a trained signal estimation model to perform signal estimation on the signal to obtain a target signal; wherein the signal estimation model is trained using the device described in claim 9.

11. An electronic device, characterized in that: The method 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 method according to any one of claims 1 to 6 is implemented, or the method according to any one of claims 7 to 8 is implemented.

12. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented, or the method according to any one of claims 7 to 8 is implemented.

13. A chip, characterized in that: The chip comprises a processing circuit, and the processing circuit is configured to execute the method according to any one of claims 1-6, or to implement the method according to any one of claims 7-8.

14. A computer program product, characterized in that The invention comprises a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 6, or implements the method according to any one of claims 7 to 8.

Images