A self-interference cancellation method based on ResRNN network

Abstract

The present application provides a self-interference cancellation method based on a ResRNN network, the method comprising: determining input data based on the real part and imaginary part of complex data. Processing the input data through the input layer to obtain a two-dimensional graph of the input data. Performing feature extraction on the two-dimensional graph through a convolution layer to obtain a feature extraction result, and processing the two-dimensional graph through a jump connection layer to obtain a first output result. Reshaping the feature extraction result through a data reshaping layer to obtain a reshaping result. Processing the reshaping result through a recursive network layer to obtain a recursive result. Processing the recursive result through an output layer to obtain a second output result. Outputting a self-interference cancellation prediction result based on the first output result and the second output result. The method provided in the present application has a relatively excellent self-interference cancellation effect while ensuring a relatively small network scale.

Description

Technical Field

[0001] The present application relates to the field of communication technology, and in particular to a self-interference cancellation method based on a ResRNN network. Background Art

[0002] With the continuous development of science and technology, wireless communication technology has been applied to all aspects of people's lives, which has led to the increasing shortage of spectrum resources used by wireless communication technology. Although traditional time division duplex (TDD) and frequency division duplex (FDD) technologies are widely used, they can only communicate at different times or frequency bands, resulting in low spectrum utilization. Full-duplex communication technology, as a solution to improve spectrum utilization, allows communication devices to send and receive data at the same time and frequency band, thereby improving spectrum utilization efficiency.

[0003] However, there is a self-interference problem in full-duplex communication. Since the locally transmitted signal leaks to the receiving end through paths such as antenna coupling and reflection, a strong self-interference signal is generated. The strength of the self-interference signal often far exceeds the desired signal at the far end, which seriously reduces the quality of the received signal and causes a sharp drop in the signal-to-noise ratio (SNR), thus affecting the communication performance. Therefore, self-interference cancellation has become a key technical challenge in full-duplex communication systems. In recent years, deep learning technology has been introduced into the field of communications, especially for self-interference cancellation. Fully connected neural networks perform well in fitting nonlinear interference signals, but the network usually requires a large parameter scale and is difficult to be directly applied to actual communication systems.

[0004] Therefore, how to improve the self-interference cancellation effect while ensuring a small network scale has become an urgent problem to be solved by those skilled in the art. Summary of the invention

[0005] Based on the above problems, the present application provides a self-interference cancellation method based on the ResRNN network to improve the self-interference cancellation effect while ensuring a small network scale.

[0006] The present application provides a self-interference cancellation method based on a ResRNN network, the method comprising:

[0007] determining input data based on the real and imaginary parts of the complex data transmitted in the digital baseband of the communication device;

[0008] Inputting the input data into an input layer, and processing the input data through the input layer to obtain a two-dimensional graph of the input data;

[0009] Inputting the two-dimensional graph into a skip connection layer and a convolution layer, performing feature extraction on the two-dimensional graph through the convolution layer to obtain a feature extraction result, and processing the two-dimensional graph through the skip connection layer to obtain a first output result;

[0010] Inputting the feature extraction result into a data reshaping layer, and reshaping the feature extraction result through the data reshaping layer to obtain a reshaping result;

[0011] Inputting the reshaped result into a recursive network layer, and processing the reshaped result through the recursive network layer to obtain a recursive result;

[0012] Inputting the recursive result into an output layer, and processing the recursive result through the output layer to obtain a second output result;

[0013] A self-interference cancellation prediction result is output based on the first output result and the second output result.

[0014] In a possible implementation, the two-dimensional graph of the input data is represented by the following formula:

[0015] = ;

[0016] in, represents input data, I represents the imaginary part of the complex data, Q represents the real part of the complex data, The [M, 2]-dimensional matrix represents component I and component Q, where M represents a delay coefficient and n represents the nth complex data.

[0017] In a possible implementation, extracting features from the two-dimensional image through the convolution layer to obtain a feature extraction result is performed by the following formula:

[0018] ;

[0019] in, represents the feature extraction result of the lth filter, N is an integer less than M, l represents the number of filters set in the convolution layer, I represents the imaginary part of the complex data, Q represents the real part of the complex data, i is an integer not greater than l, and the data dimension of the feature extraction result is , k represents the parameters of the convolutional layer.

[0020] In a possible implementation manner, the processing the two-dimensional graph through the skip connection layer to obtain a first output result includes:

[0021] Set the network parameters of the jump connection layer to Linear (M, 2), where M represents a delay coefficient;

[0022] The first output result is obtained by processing the two-dimensional graph through the jump connection layer after setting the parameters .

[0023] In a possible implementation, reshaping the feature extraction result by the data reshaping layer to obtain the reshaping result is performed by the following formula:

[0024] ];

[0025] in, represents the reshaping result, Respectively represent The feature extraction result of the filter output is The data dimension is .

[0026] In a possible implementation, the processing of the reshaping result by the recursive network layer to obtain the recursive result is performed by the following formula:

[0027] ;

[0028] in, represents the recursive result of the loop layer output at any time step t, ∈ , represents the number of neurons in the recurrent layer, and R represents the parameters of the recurrent network layer;

[0029] represents the output of the recurrent layer at the previous time step t-1, ∈ ;

[0030] express A row vector of represents the reshaping result, Passed to the recurrent layer at time step t, ∈ , Indicates the number of input features;

[0031] represents the weight matrix of the connection between the input unit and the hidden unit in the current time step t, ∈ ;

[0032] represents the weight matrix of the feedback connection from the hidden unit in the previous time step t-1, ∈ ;

[0033] (.) represents the activation function operation of the loop layer;

[0034] represents the row vector containing the bias terms of the neurons in the recurrent layer, ∈ .

[0035] In a possible implementation, the processing of the recursive result by the output layer to obtain the second output result is performed by the following formula:

[0036] = ;

[0037] in, Represents the second output result, Represents the imaginary part output result, Represents the real part output result;

[0038] ;

[0039] = ;

[0040] in, (.) represents the activation function operation of the output layer, represents the weight and bias term associated with the first neuron in the output layer, represents the weight and bias term associated with the second neuron in the output layer, represents the i-th recursive result, Represents the number of neurons in the recurrent layer.

[0041] In a possible implementation manner, outputting the self-interference cancellation prediction result based on the first output result and the second output result is performed by the following formula:

[0042] ;

[0043] in, represents the self-interference cancellation prediction result, Represents the second output result, Indicates the first output result.

[0044] The present application also provides a self-interference cancellation device based on a ResRNN network, the device comprising the following modules:

[0045] a determination module for determining input data based on the real and imaginary parts of the complex data transmitted in the digital baseband of the communication device;

[0046] A first input module, used for inputting the input data into an input layer, and processing the input data through the input layer to obtain a two-dimensional graph of the input data;

[0047] A second input module is used to input the two-dimensional graph into a skip connection layer and a convolution layer, perform feature extraction on the two-dimensional graph through the convolution layer to obtain a feature extraction result, and process the two-dimensional graph through the skip connection layer to obtain a first output result;

[0048] A third input module is used to input the feature extraction result into a data reshaping layer, and reshape the feature extraction result through the data reshaping layer to obtain a reshaping result;

[0049] A fourth input module, used for inputting the reshaped result into the recursive network layer, and processing the reshaped result through the recursive network layer to obtain a recursive result;

[0050] A fifth input module, used for inputting the recursive result into an output layer, and processing the recursive result through the output layer to obtain a second output result;

[0051] An output module is used to output a self-interference cancellation prediction result based on the first output result and the second output result.

[0052] The present application also provides a ResRNN network for self-interference cancellation, the network comprising:

[0053] Input layer, convolutional layer, skip connection layer, data reshaping layer, recurrent network layer, and output layer;

[0054] The input layer is connected to the convolution layer, the input layer is used to process input data to obtain a two-dimensional graph of the input data, the convolution layer includes a plurality of filters, and the convolution layer is used to perform feature extraction on the two-dimensional graph to obtain a feature extraction result;

[0055] The input layer is connected to the jump connection layer, and the jump connection layer is used to process the two-dimensional graph to obtain a first output result;

[0056] The convolution layer is connected to the data reshaping layer, and the data reshaping layer is used to reshape the feature extraction result to obtain a reshaping result;

[0057] The data reshaping layer is connected to the recursive network layer, and the recursive network layer is used to process the reshaping result to obtain a recursive result;

[0058] The recursive network layer is connected to the output layer, and the output layer is used to process the recursive result to obtain a second output result.

[0059] The present application also provides an electronic device, the electronic device comprising a processor and a memory:

[0060] The memory is used to store a computer program and transmit the computer program to the processor;

[0061] The processor is used to execute the steps of the above-mentioned self-interference cancellation method based on the ResRNN network according to the instructions in the computer program.

[0062] The present application also provides a computer-readable storage medium, which is used to store a computer program. When the computer program is executed by an electronic device, the steps of the above-mentioned self-interference cancellation method based on the ResRNN network are implemented.

[0063] Compared with the prior art, this application has the following beneficial effects:

[0064] Compared with the prior art of the self-interference cancellation method based on ResRNN network provided by the present application, the present application introduces residual connection and recursive neural network, and effectively reduces the parameter scale by combining residual connection and recursive neural network. While reducing the parameter scale, the present application extracts features of the input data through the convolution layer, which not only effectively retains the key information in the input data, but also extracts spatial features through convolution operation, providing more accurate input data for subsequent self-interference cancellation. The method provided by the present application can better process the time series features in the data through the design of the data reshaping layer and the recursive network layer, so that the self-interference cancellation is more accurate. When processing complex data, the method can be optimized according to the actual signal characteristics, avoiding the overfitting or underfitting problems that may occur in the traditional method, and improving the robustness and adaptability of the system. Based on the first output result and the second output result, the self-interference cancellation prediction result is output, so that the self-interference cancellation prediction result can retain more features of the original data, alleviate the problem of model learning ability degradation, and enhance the ability of model learning nonlinearity. While reducing the parameter scale, the present application has a relatively excellent self-interference cancellation effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0066] Figure 1 A schematic diagram of the structure of a ResRNN network for self-interference cancellation provided in an embodiment of the present application;

[0067] Figure 2 A flowchart of a self-interference cancellation method based on a ResRNN network provided in an embodiment of the present application;

[0068] Figure 3 A schematic diagram of the structure of a self-interference cancellation device based on a ResRNN network provided in an embodiment of the present application. DETAILED DESCRIPTION

[0069] As described above, in recent years, deep learning technology has been introduced into the field of communications, especially for self-interference cancellation. Fully connected neural networks perform well in fitting nonlinear interference signals, but the network usually requires a large parameter scale and is difficult to be directly applied to actual communication systems.

[0070] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0071] It is understandable that the method provided in the present application can be applied to a processing device, which is a processing device that can determine input data based on the real part and imaginary part of the complex data sent in the digital baseband of the communication device, for example, a terminal device or server that can determine input data based on the real part and imaginary part of the complex data sent in the digital baseband of the communication device. The method provided in the present application can be executed independently by a terminal device or a server, and can also be applied to a network scenario in which a terminal device and a server communicate, and is executed by the cooperation of a terminal device and a server. Among them, the terminal device can be a computer, a mobile phone and other devices. The server can be understood as an application server or a Web server. In actual deployment, the server can be a stand-alone server or a cluster server.

[0072] Figure 1 A schematic diagram of the structure of a ResRNN network for self-interference cancellation provided in an embodiment of the present application, Figure 1 The ResRNN network for self-interference cancellation in includes an input layer, a convolutional layer, a skip connection layer, a data reshaping layer, a recursive network layer and an output layer.

[0073] Understandably, Figure 1Only for the purpose of explaining the ResRNN network structure used for self-interference cancellation, Figure 1 The parameters appearing in will be described in detail later.

[0074] The input layer is connected to the convolution layer, and the input layer is used to process the input data to obtain a two-dimensional graph of the input data. The data scale of the two-dimensional graph obtained by the input layer can be 13x2.

[0075] The convolution layer includes multiple filters, and the convolution layer is used to extract features from the two-dimensional image to obtain feature extraction results.

[0076] The input layer is connected to the skip connection layer, and the skip connection layer is used to process the two-dimensional graph to obtain a first output result.

[0077] The convolution layer is connected to the data reshaping layer, and the data reshaping layer is used to reshape the feature extraction result to obtain a reshaped result.

[0078] The data reshaping layer is connected to the recursive network layer, and the recursive network layer is used to process the reshaping results to obtain recursive results.

[0079] The recursive network layer is connected to the output layer, and the output layer is used to process the recursive result to obtain a second output result. and Represents the two components of the second output result.

[0080] The final self-interference cancellation prediction result is obtained through the first output result output by the jump connection layer and the second output result output by the output layer.

[0081] Figure 2 A flowchart of a self-interference cancellation method based on a ResRNN network provided in an embodiment of the present application, the method comprising the following steps:

[0082] S101: Determine input data based on the real part and the imaginary part of the complex data transmitted in the digital baseband of the communication device.

[0083] The processing device determines input data based on the real and imaginary parts of the complex data transmitted in the digital baseband of the communication device.

[0084] In digital communication systems, complex data is achieved by representing the signal as a combination of real and imaginary parts, which is mathematically represented as z = a + bi, where a is the real part and b is the imaginary part. This representation provides us with the ability to plot the amplitude and phase changes of the signal on the complex plane, which is very useful for signal processing. For example, when we use IQ modulation, a signal is decomposed into two mutually orthogonal components, the I (In-phase) component and the Q (Quadrature) component.

[0085] In a possible implementation, components I and Q obtained by IQ-modulating the real part and the imaginary part of the complex data sent in the digital baseband of the communication device in full-duplex communication can be used as input data.

[0086] S102: Inputting input data into an input layer, and processing the input data through the input layer to obtain a two-dimensional graph of the input data.

[0087] The processing device inputs the input data into the input layer, and processes the input data through the input layer to obtain a two-dimensional graph of the input data.

[0088] In a possible implementation, the input layer can represent the input data as a two-dimensional graph. The two-dimensional graph of the input data can be represented by the following formula:

[0089] = ;

[0090] in, represents input data, I represents the imaginary part of the complex data, Q represents the real part of the complex data, The [M, 2]-dimensional matrix represents component I and component Q, where M represents a delay coefficient and n represents the nth complex data.

[0091] In the actual application process, M needs to be evaluated according to the actual RF link situation. After a lot of experimental research, M in this network can be set to 13, which is a better choice. In this case, the input data is [13,2]-dimensional I / Q data.

[0092] S103: Input the two-dimensional image into the skip connection layer and the convolution layer, perform feature extraction on the two-dimensional image through the convolution layer to obtain a feature extraction result, and process the two-dimensional image through the skip connection layer to obtain a first output result.

[0093] The processing device can input the two-dimensional image into the skip connection layer and the convolution layer, extract features of the two-dimensional image through the convolution layer to obtain a feature extraction result, and process the two-dimensional image through the skip connection layer to obtain a first output result.

[0094] In a possible implementation, the convolutional layer is the first hidden layer of the network, and the convolutional layer can use Conv2d convolution. In order to reduce the scale and computational complexity of the network, the convolution kernel size can be set to N×1, where N<M, and as large as possible within the range of N<M. The number of filters in the convolutional layer is set to After a lot of research, we found that when N=9, =3 when the self-interference cancellation effect is the best.

[0095] In a possible implementation, the feature extraction result obtained by performing feature extraction on the two-dimensional image through the convolution layer can be obtained by the following formula:

[0096] ;

[0097] in, represents the feature extraction result of the lth filter, N is an integer less than M, l represents the number of filters set in the convolution layer, I represents the imaginary part of the complex data, Q represents the real part of the complex data, i is an integer not greater than l, and the data dimension of the feature extraction result is , k represents the parameters of the convolutional layer.

[0098] In a possible implementation, the skip connection layer and the convolution layer are parallel structures, and are also the first hidden network layer of the network. The input layer data is input into the convolution layer and the skip connection layer at the same time.

[0099] In a possible implementation, the first output result can be obtained by processing the two-dimensional graph through the jump connection layer. The network parameters of the jump connection layer can be set to Linear (M, 2), where M represents the delay coefficient. Then, the first output result is obtained by processing the two-dimensional graph through the jump connection layer after setting the parameters. .

[0100] The skip connection layer adopts a residual structure. The skip connection layer skips the data reshaping layer and the recursive neural network layer and is directly connected to the output layer, so the network parameters can be set to Linear (M, 2).

[0101] The method provided in this application can retain more features of the original data through the skip connection layer, alleviate the problem of model learning ability degradation, and enhance the model's ability to learn nonlinearity.

[0102] For example, when the data of the input layer is [13,2] dimensional data, the data can be respectively (8×1), (8×1), (8×1) performs convolution operation and obtains dimensional vector data.

[0103] S104: Input the feature extraction result into the data reshaping layer, and reshape the feature extraction result through the data reshaping layer to obtain a reshaping result.

[0104] The processing device can input the feature extraction result into the data reshaping layer, and reshape the feature extraction result through the data reshaping layer to obtain a reshaping result.

[0105] In a possible implementation, reshaping the feature extraction result through the data reshaping layer to obtain the reshaping result can be performed by the following formula:

[0106] ];

[0107] in, represents the reshaping result, Respectively represent The feature extraction result of the filter output is The data dimension is .

[0108] In order to more effectively obtain the temporal relationship of the data, the input data needs to be reshaped before being passed to the recurrent network layer after feature extraction through the filter of the convolutional layer. Since there is temporal behavior in the input data, there is also a dependency between each column element of the feature map output by each filter. In order to better advance this dependency, the above formula can be used to reshape them and pass them to the recurrent network layer in sequence, so that the recurrent network layer can detect the above features and appropriately model the temporal behavior of the data.

[0109] For example, the data reshaping layer can transform the output of the convolutional layer into (6×2), (6×2), (6×2) Data reshaping layer is formed according to the row dimension (6×6) dimensional data, used as input data for the recurrent network layer.

[0110] S105: Input the reshaped result into the recursive network layer, and process the reshaped result through the recursive network layer to obtain a recursive result.

[0111] The processing device can input the reshaped result into the recursive network layer, and process the reshaped result through the recursive network layer to obtain the recursive result.

[0112] In one possible implementation, the recursive result obtained by processing the reshaped result through the recursive network layer can be obtained by the following formula:

[0113] ;

[0114] in, represents the recursive result of the loop layer output at any time step t, ∈ , represents the number of neurons in the recurrent layer, and R represents the parameters of the recurrent network layer;

[0115] represents the output of the recurrent layer at the previous time step t-1, ∈ ;

[0116] express A row vector of represents the reshaping result, Passed to the recurrent layer at time step t, ∈ , Indicates the number of input features;

[0117] represents the weight matrix of the connection between the input unit and the hidden unit in the current time step t, ∈ ;

[0118] represents the weight matrix of the feedback connection from the hidden unit in the previous time step t-1, ∈ ;

[0119] (.) represents the activation function operation of the loop layer;

[0120] represents the row vector containing the bias terms of the neurons in the recurrent layer, ∈ .

[0121] The reshaped result after reshaping Passed to the recurrent network layer, the output at any time step t can be expressed as .

[0122] S106: Input the recursive result to the output layer, and process the recursive result through the output layer to obtain a second output result.

[0123] The processing device inputs the recursive result into the output layer, and processes the recursive result through the output layer to obtain a second output result.

[0124] In a possible implementation, the second output structure of the output layer output can be implemented by a fully connected network with two outputs, and the second output result is the I / Q component of the nonlinear interference at the current time t in the predicted communication channel, and the expression formula is as follows:

[0125] = ;

[0126] in, Represents the second output result, Represents the imaginary part output result, Represents the real part output result;

[0127] ;

[0128] = ;

[0129] in, (.) represents the activation function operation of the output layer, represents the weight and bias term associated with the first neuron in the output layer, represents the weight and bias term associated with the second neuron in the output layer, represents the i-th recursive result, Represents the number of neurons in the recurrent layer.

[0130] S107: Output the self-interference cancellation prediction result based on the first output result and the second output result.

[0131] The processing device outputs a self-interference cancellation prediction result based on the first output result and the second output result.

[0132] In a possible implementation, outputting the self-interference cancellation prediction result based on the first output result and the second output result may be performed by the following formula:

[0133] ;

[0134] in, represents the self-interference cancellation prediction result, Represents the second output result, Indicates the first output result.

[0135] Compared with the prior art of the self-interference cancellation method based on ResRNN network provided by the present application, the present application introduces residual connection and recursive neural network, and effectively reduces the parameter scale by combining residual connection and recursive neural network. While reducing the parameter scale, the present application extracts features of the input data through the convolution layer, which not only effectively retains the key information in the input data, but also extracts spatial features through convolution operation, providing more accurate input data for subsequent self-interference cancellation. The method provided by the present application can better process the time series features in the data through the design of the data reshaping layer and the recursive network layer, so that the self-interference cancellation is more accurate. When processing complex data, the method can be optimized according to the actual signal characteristics, avoiding the overfitting or underfitting problems that may occur in the traditional method, and improving the robustness and adaptability of the system. Based on the first output result and the second output result, the self-interference cancellation prediction result is output, so that the self-interference cancellation prediction result can retain more features of the original data, alleviate the problem of model learning ability degradation, and enhance the ability of model learning nonlinearity. While reducing the parameter scale, the present application has a relatively excellent self-interference cancellation effect.

[0136] The present application also provides a Figure 3 The structure diagram of the self-interference cancellation device based on the ResRNN network is shown, and the self-interference cancellation device 200 based on the ResRNN network includes:

[0137] A determination module 201 for determining input data based on the real part and the imaginary part of the complex data transmitted in the digital baseband of the communication device;

[0138] A first input module 202, used for inputting the input data into an input layer, and processing the input data through the input layer to obtain a two-dimensional graph of the input data;

[0139] The second input module 203 is used to input the two-dimensional graph into the skip connection layer and the convolution layer, perform feature extraction on the two-dimensional graph through the convolution layer to obtain a feature extraction result, and process the two-dimensional graph through the skip connection layer to obtain a first output result;

[0140] A third input module 204 is used to input the feature extraction result into a data reshaping layer, and reshape the feature extraction result through the data reshaping layer to obtain a reshaping result;

[0141] A fourth input module 205, used for inputting the reshaped result into the recursive network layer, and processing the reshaped result through the recursive network layer to obtain a recursive result;

[0142] A fifth input module 206, configured to input the recursive result into an output layer, and process the recursive result through the output layer to obtain a second output result;

[0143] The output module 207 is configured to output a self-interference cancellation prediction result based on the first output result and the second output result.

[0144] Compared with the prior art of the self-interference cancellation device based on the ResRNN network provided by the present application, the present application introduces residual connection and recursive neural network, and effectively reduces the parameter scale by combining residual connection and recursive neural network. While reducing the parameter scale, the present application extracts features of the input data through the convolution layer, which not only effectively retains the key information in the input data, but also extracts spatial features through the convolution operation, providing more accurate input data for subsequent self-interference cancellation. The method provided by the present application can better process the time series features in the data through the design of the data reshaping layer and the recursive network layer, so that the self-interference cancellation is more accurate. When processing complex data, the method can be optimized according to the actual signal characteristics, avoiding the overfitting or underfitting problems that may occur in the traditional method, and improving the robustness and adaptability of the system. Based on the first output result and the second output result, the self-interference cancellation prediction result is output, so that the self-interference cancellation prediction result can retain more features of the original data, alleviate the problem of model learning ability degradation, and enhance the model learning nonlinear ability. While reducing the parameter scale, the present application has a relatively excellent self-interference cancellation effect.

[0145] An embodiment of the present application also provides a self-interference cancellation device based on a ResRNN network, wherein the device includes a memory and a processor, the memory is used to store instructions or codes, and the processor is used to execute the instructions or codes so that the device performs the steps of the self-interference cancellation method based on the ResRNN network described in any embodiment of the present application.

[0146] In practical applications, the computer-readable storage medium may be any combination of one or more computer-readable media, which may be a computer-readable signal medium or a computer-readable storage medium.

[0147] The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, device, or device.

[0148] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0149] The program code embodied on the computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0150] Computer program code for performing the operations of the present invention may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0151] It should be noted that each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. The device embodiment described above is merely schematic, in which the unit described as a separate component may or may not be physically separated, and the component prompted as a unit may or may not be a physical unit, that is, it may be located in one place, or it may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative work.

[0152] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A self-interference cancellation method based on ResRNN network, characterized in that: include: determining input data based on the real and imaginary parts of the complex data transmitted in the digital baseband of the communication device; Inputting the input data into an input layer, and processing the input data through the input layer to obtain a two-dimensional graph of the input data; Inputting the two-dimensional graph into a skip connection layer and a convolution layer, performing feature extraction on the two-dimensional graph through the convolution layer to obtain a feature extraction result, and processing the two-dimensional graph through the skip connection layer to obtain a first output result; Inputting the feature extraction result into a data reshaping layer, and reshaping the feature extraction result through the data reshaping layer to obtain a reshaping result; The reshaping of the feature extraction result by the data reshaping layer to obtain the reshaping result is performed by the following formula: ]； in, represents the reshaping result, Respectively represent The feature extraction result of the filter output is The data dimension is , M represents the delay coefficient, N is an integer less than M, Represents the number of filters set in the convolutional layer; Inputting the reshaped result into a recursive network layer, and processing the reshaped result through the recursive network layer to obtain a recursive result; The reshaping result is processed by the recursive network layer to obtain the recursive result by the following formula: ； in, represents the recursive result of the loop layer output at any time step t, ∈ , represents the number of neurons in the recurrent layer, and R represents the parameters of the recurrent network layer; represents the output of the recurrent layer at the previous time step t-1, ∈ ; express A row vector of represents the reshaping result, Passed to the recurrent layer at time step t, ∈ , Indicates the number of input features; represents the weight matrix of the connection between the input unit and the hidden unit in the current time step t, ∈ ; represents the weight matrix of the feedback connection from the hidden unit in the previous time step t-1, ∈ ; (.) represents the activation function operation of the loop layer; represents the row vector containing the bias terms of the neurons in the recurrent layer, ∈ ; Inputting the recursive result into an output layer, and processing the recursive result through the output layer to obtain a second output result; A self-interference cancellation prediction result is output based on the first output result and the second output result.

2. The method according to claim 1, characterized in that The two-dimensional graph of the input data is represented by the following formula: = ； in, represents input data, I represents the imaginary part of the complex data, Q represents the real part of the complex data, The [M, 2]-dimensional matrix represents the components I and Q, and n represents the n-th complex data.

3. The method according to claim 2, characterized in that The feature extraction result obtained by performing feature extraction on the two-dimensional image through the convolution layer is performed by the following formula: ； in, represents the feature extraction result of the lth filter, I represents the imaginary part of the complex data, Q represents the real part of the complex data, i is an integer not greater than l, and the data dimension of the feature extraction result is , k represents the parameters of the convolutional layer.

4. The method according to claim 3, characterized in that The processing of the two-dimensional graph by the skip connection layer to obtain a first output result includes: Set the network parameters of the jump connection layer to Linear (M, 2), where M represents a delay coefficient; The first output result is obtained by processing the two-dimensional graph through the jump connection layer after setting the parameters .

5. The method according to claim 3, characterized in that: The processing of the recursive result by the output layer to obtain the second output result is performed by the following formula: = ； in, Represents the second output result, Represents the imaginary output result, Represents the real part output result; ； = ； in, (.) represents the activation function operation of the output layer, represents the weight and bias term associated with the first neuron in the output layer, represents the weight and bias term associated with the second neuron in the output layer, represents the i-th recursive result, Represents the number of neurons in the recurrent layer.

6. The method according to claim 5, characterized in that The outputting of the self-interference cancellation prediction result based on the first output result and the second output result is performed by the following formula: ； in, represents the self-interference cancellation prediction result, Represents the second output result, Indicates the first output result.

7. A self-interference cancellation device based on ResRNN network, characterized in that: include: a determination module for determining input data based on the real and imaginary parts of the complex data transmitted in the digital baseband of the communication device; A first input module, used for inputting the input data into an input layer, and processing the input data through the input layer to obtain a two-dimensional graph of the input data; A second input module is used to input the two-dimensional graph into a skip connection layer and a convolution layer, perform feature extraction on the two-dimensional graph through the convolution layer to obtain a feature extraction result, and process the two-dimensional graph through the skip connection layer to obtain a first output result; A third input module is used to input the feature extraction result into a data reshaping layer, and reshape the feature extraction result through the data reshaping layer to obtain a reshaping result; The reshaping of the feature extraction result by the data reshaping layer to obtain the reshaping result is performed by the following formula: ]； in, represents the reshaping result, Respectively represent The feature extraction result of the filter output is The data dimension is , M represents the delay coefficient, N is an integer less than M, Represents the number of filters set in the convolutional layer; A fourth input module, used for inputting the reshaped result into the recursive network layer, and processing the reshaped result through the recursive network layer to obtain a recursive result; The reshaping result is processed by the recursive network layer to obtain the recursive result by the following formula: ； in, represents the recursive result of the loop layer output at any time step t, ∈ , represents the number of neurons in the recurrent layer, and R represents the parameters of the recurrent network layer; represents the output of the recurrent layer at the previous time step t-1, ∈ ; express A row vector of represents the reshaping result, Passed to the recurrent layer at time step t, ∈ , Indicates the number of input features; represents the weight matrix of the connection between the input unit and the hidden unit in the current time step t, ∈ ; represents the weight matrix of the feedback connection from the hidden unit in the previous time step t-1, ∈ ; (.) represents the activation function operation of the loop layer; represents the row vector containing the bias terms of the neurons in the recurrent layer, ∈ ; A fifth input module, used for inputting the recursive result into an output layer, and processing the recursive result through the output layer to obtain a second output result; An output module is used to output a self-interference cancellation prediction result based on the first output result and the second output result.

8. A ResRNN network for self-interference cancellation, characterized in that: include: Input layer, convolutional layer, skip connection layer, data reshaping layer, recurrent network layer, and output layer; The input layer is connected to the convolution layer, the input layer is used to process input data to obtain a two-dimensional graph of the input data, the convolution layer includes a plurality of filters, and the convolution layer is used to perform feature extraction on the two-dimensional graph to obtain a feature extraction result; The input layer is connected to the jump connection layer, and the jump connection layer is used to process the two-dimensional graph to obtain a first output result; The convolution layer is connected to the data reshaping layer, and the data reshaping layer is used to reshape the feature extraction result to obtain a reshaping result; The reshaping of the feature extraction result by the data reshaping layer to obtain the reshaping result is performed by the following formula: ]； in, represents the reshaping result, Respectively represent The feature extraction result of the filter output is The data dimension is , M represents the delay coefficient, N is an integer less than M, Represents the number of filters set in the convolutional layer; The data reshaping layer is connected to the recursive network layer, and the recursive network layer is used to process the reshaping result to obtain a recursive result; The reshaping result is processed by the recursive network layer to obtain the recursive result by the following formula: ； in, represents the recursive result of the loop layer output at any time step t, ∈ , represents the number of neurons in the recurrent layer, and R represents the parameters of the recurrent network layer; represents the output of the recurrent layer at the previous time step t-1, ∈ ; express A row vector of represents the reshaping result, Passed to the recurrent layer at time step t, ∈ , Indicates the number of input features; represents the weight matrix of the connection between the input unit and the hidden unit in the current time step t, ∈ ; represents the weight matrix of the feedback connection from the hidden unit in the previous time step t-1, ∈ ; (.) represents the activation function operation of the loop layer; represents the row vector containing the bias terms of the neurons in the recurrent layer, ∈ ; The recursive network layer is connected to the output layer, and the output layer is used to process the recursive result to obtain a second output result.

Images