A digital self-interference cancellation method and device based on an enhanced neural network structure

Abstract

The application discloses a digital self-interference cancellation method and device based on an enhanced neural network structure, and the method comprises the following steps: performing complex value splitting operation and non-linearization operation on a first ladder grid structure to obtain a second ladder grid structure, and performing complex value splitting operation and non-linearization operation on a first sliding window grid structure to obtain a second sliding window grid structure; obtaining a first output of a dynamic phase layer according to the second ladder grid structure or the second sliding window grid structure; constructing a vector decomposition structure to separate amplitude information and phase information; feeding the amplitude information into a non-linear hidden layer to obtain a second output; fusing the phase information and the second output; feeding the fused data into a full connection output layer to output a simulation result; and performing self-interference cancellation operation on a received signal according to the simulation result to obtain a target signal. The application can achieve high digital self-interference cancellation performance and reduce network complexity, and can be widely applied to the field of full-duplex communication technology.

Description

Technical Field

[0001] This invention relates to the field of full-duplex communication technology, and in particular to a digital self-interference cancellation method and apparatus based on an enhanced neural network structure. Background Technology

[0002] Most existing neural network (NN) architectures suffer from high computational complexity and poor self-interference (SI) suppression. While typical NNs can suppress nonlinear SI to some extent when modeling it, the process of simulating nonlinearity in NNs does not conform to the actual generation patterns of nonlinear SI, necessitating high computational complexity to ensure cancellation performance. Existing improved NNs primarily reduce network complexity by optimizing typical NN structures or cascading different typical NNs. While achieving some success, these improvements still fail to address the generation patterns of nonlinear SI, resulting in persistently high computational complexity. Summary of the Invention

[0003] In view of this, the main objective of the embodiments of the present invention is to provide a digital self-interference cancellation method and apparatus based on an enhanced neural network structure, in order to solve at least one of the problems of the prior art. The present invention can improve the performance of digital self-interference cancellation and reduce network complexity.

[0004] To achieve the above objectives, one aspect of the present invention provides a digital self-interference cancellation method based on an enhanced neural network structure, comprising the following steps:

[0005] The first step mesh structure is subjected to complex value decomposition and nonlinearity removal operations to obtain the second step mesh structure. The first sliding window mesh structure is subjected to complex value decomposition and nonlinearity removal operations to obtain the second sliding window mesh structure.

[0006] Based on the second stepped mesh structure or the second sliding window mesh structure, obtain the first output of the dynamic phase layer;

[0007] A vector decomposition structure is constructed, and amplitude information and phase information are separated based on the first output of the dynamic phase layer and the vector decomposition structure.

[0008] The amplitude information is fed into the nonlinear hidden layer to obtain the second output of the nonlinear hidden layer;

[0009] The phase information and the second output are fused to obtain fused data;

[0010] The fused data is fed into a fully connected output layer to output simulation results;

[0011] Based on the simulation results, a self-interference cancellation operation is performed on the received signal to obtain the target signal.

[0012] In some embodiments, a digital self-interference cancellation method based on an enhanced neural network structure further includes the following steps:

[0013] Obtain the full-duplex transmit and receive data set;

[0014] The full-duplex transmit and receive data set is split to obtain a training set and a test set;

[0015] Based on the training set, the first neural network based on the second ladder grid structure is trained to obtain the second neural network;

[0016] Based on the training set, the third neural network based on the second sliding window mesh structure is trained to obtain the fourth neural network;

[0017] Based on the test set, the self-interference cancellation performance of the second neural network and the fourth neural network is tested, and the test results are obtained.

[0018] In some embodiments, performing complex value decomposition and nonlinearization operations on the first stepped mesh structure to obtain the second stepped mesh structure includes the following steps:

[0019] The first complex value of the first stepped mesh structure is split into a first real part and a first imaginary part;

[0020] A first neuron is added to the first hidden layer of the first stepped grid structure;

[0021] The original second neuron of the first step grid structure is combined with the first neuron, and the initial activation function of the first hidden layer is replaced with a linear activation function to obtain the second hidden layer;

[0022] The second stepped mesh structure is obtained based on the first real part, the first imaginary part, and the second hidden layer.

[0023] In some embodiments, performing complex value decomposition and nonlinearity removal operations on the first sliding window mesh structure to obtain the second sliding window mesh structure includes the following steps:

[0024] The second complex value of the first sliding window mesh structure is split into a second real part and a second imaginary part;

[0025] A third neuron is added to the third hidden layer of the first sliding window mesh structure;

[0026] The original fourth neuron of the first sliding window mesh structure is combined with the third neuron, and the initial activation function of the third hidden layer is replaced with a linear activation function to obtain the fourth hidden layer.

[0027] The second sliding window mesh structure is obtained based on the second real part, the second imaginary part, and the fourth hidden layer.

[0028] In some embodiments, the formula used to obtain the first output of the dynamic phase layer based on the second stepped mesh structure or the second sliding window mesh structure includes:

[0029] o ′ (n)=[o ′ 1(n),o ′ 2(n),…,o ′ 2G-1 (n),o ′ 2G (n)] T ;

[0030] Among them, o ′ (n) represents the first output of the dynamic phase layer; o ′ 1(n),o ′ 2(n),…,o ′ 2G-1 (n),o ′ 2G (n) represents the output of each neuron in the dynamic phase layer; G represents the number of neurons in the nonlinear hidden layer; n represents the discrete time step; [·] T This indicates the transpose operation.

[0031] In some embodiments, the formula used to construct the vector decomposition structure, separating amplitude information and phase information based on the first output of the dynamic phase layer and the vector decomposition structure, includes:

[0032]

[0033] Among them, s i (n) represents amplitude information; sinθ i Cosθ represents the phase of a sinusoidal wave. i G represents the cosine phase; G represents the number of neurons in the nonlinear hidden layer; n represents the discrete time step; o ′ 2i (n), o ′ 2i-1 (n) represents the output of the 2i-th and 2i-1-th neurons in the dynamic phase layer.

[0034] In some embodiments, the formula used to feed the amplitude information into the nonlinear hidden layer to obtain the second output of the nonlinear hidden layer includes:

[0035] x ′ NN (n)=[|s1(n)|,|s1(n)| 2 ,…,|s G (n)|,|s G (n)| 2 ] T ;

[0036] o ″ NN (n)=[o1 ″ (n),o2 ″ (n),…,o G ″ (n)] T ;

[0037] Where, x ′ NN (n) represents the input of the nonlinear hidden layer; s1(n),…,s G (n) represents amplitude information; [·] T Indicates the transpose operation; o ″ NN (n) represents the second output of the nonlinear hidden layer; o1 ″ (n),o2 ″ (n),…,o G ″ (n) represents the output of each neuron in the nonlinear hidden layer.

[0038] In some embodiments, the formula used to perform the fusion operation on the phase information and the second output to obtain fused data includes:

[0039] x ″ NN (n)=[o1 ″ (n)sinθ1,o1 ′ (n)cosθ1,…,o G ″ (n)sinθ G ,o G ″ (n)cosθ G ] T ;

[0040] Where, x ″ NN (n) represents the fused data; o1 ″ (n),…,o G″ (n) represents the output of each neuron in the nonlinear hidden layer; sinθ1,…,sinθ G Represents the phases of each sine wave; cosθ1,…,cosθ G Indicates the phase of each cosine; [·] T This indicates the transpose operation.

[0041] In some embodiments, the formula used to perform self-interference cancellation on the received signal based on the simulation results to obtain the target signal includes:

[0042]

[0043] Where y(n) represents the target signal; y SI (n) represents the received signal; This indicates the simulation results.

[0044] To achieve the above objectives, another aspect of the present invention proposes a digital self-interference cancellation device based on an enhanced neural network structure, the device comprising:

[0045] The first module is used to perform complex value decomposition and nonlinearization operations on the first stepped mesh structure to obtain the second stepped mesh structure, and to perform complex value decomposition and nonlinearization operations on the first sliding window mesh structure to obtain the second sliding window mesh structure.

[0046] The second module is used to obtain the first output of the dynamic phase layer based on the second stepped mesh structure or the second sliding window mesh structure;

[0047] The third module is used to construct a vector decomposition structure, and to separate amplitude information and phase information based on the first output of the dynamic phase layer and the vector decomposition structure.

[0048] The fourth module is used to feed the amplitude information into the nonlinear hidden layer to obtain the second output of the nonlinear hidden layer;

[0049] The fifth module is used to perform a fusion operation on the phase information and the second output to obtain fused data;

[0050] The sixth module is used to feed the fused data into the fully connected output layer and output the simulation results;

[0051] The seventh module is used to perform self-interference cancellation operation on the received signal based on the simulation results to obtain the target signal.

[0052] To achieve the above objectives, another aspect of the present invention provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the aforementioned digital self-interference cancellation method based on an enhanced neural network structure.

[0053] To achieve the above objectives, another aspect of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned digital self-interference cancellation method based on an enhanced neural network structure.

[0054] To achieve the above objectives, another aspect of the present invention provides a computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium and execute the computer instructions, causing the computer device to perform the aforementioned digital self-interference cancellation method based on an enhanced neural network structure.

[0055] The embodiments of the present invention include at least the following beneficial effects: The present invention provides a digital self-interference cancellation method and apparatus based on an enhanced neural network structure. This scheme obtains a second-order ... Attached Figure Description

[0056] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0057] Figure 1 This is a flowchart of a digital self-interference cancellation method based on an enhanced neural network structure provided by an embodiment of the present invention;

[0058] Figure 2 This is a schematic diagram of the complex-valued decomposition and nonlinearity removal of the LWGS structure provided in the embodiment of the present invention;

[0059] Figure 3 This is a schematic diagram of the complex-valued decomposition and nonlinearization of the MWGS structure provided in the embodiment of the present invention;

[0060] Figure 4 This is a schematic diagram of the structure of the neural network canceller based on enhanced LWGS provided in an embodiment of the present invention;

[0061] Figure 5 This is a schematic diagram of the structure of the neural network canceller based on the enhanced MWGS provided in an embodiment of the present invention;

[0062] Figure 6 This is a schematic diagram illustrating its function in a full-duplex transceiver according to an embodiment of the present invention;

[0063] Figure 7 This is a schematic diagram comparing the cancellation effects provided in the embodiments of the present invention;

[0064] Figure 8 This is a schematic diagram of the hardware structure of the electronic device provided in an embodiment of the present invention. Detailed Implementation

[0065] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this invention; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this invention as detailed in the appended claims.

[0066] It should be noted that although functional modules are divided in the system diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the system or the order in the flowchart. The terms "first / S100" and "second / S200" in the specification, claims, and the foregoing drawings may be used herein to describe various concepts, but unless specifically stated otherwise, these concepts are not limited by these terms. These terms are used only to distinguish one concept from another. For example, first information may also be referred to as second information without departing from the scope of the embodiments of the invention, and similarly, second information may also be referred to as first information. Depending on the context, the words "if" or "when" as used herein may be interpreted as "when," "in response to a determination," or "in the event of a determination."

[0067] The terms “at least one,” “multiple,” “each,” “any,” etc., used in this invention, “at least one” includes one, two, or more than two; “multiple” includes two or more than two; “each” refers to each of the corresponding multiple; and “any” refers to any one of the multiple.

[0068] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein is for the purpose of describing embodiments of the invention only and is not intended to limit the invention.

[0069] Before providing a detailed description of the embodiments of the present invention, some of the nouns and terms involved in the embodiments of the present invention will be explained first. The nouns and terms involved in the embodiments of the present invention are subject to the following interpretations.

[0070] Simultaneous Transmit and Receive (STAR): A technique that involves transmitting and receiving electromagnetic wave signals simultaneously at the same time and on the same frequency band within the same radio system.

[0071] Self-interference cancellation (SIC) refers to the process of suppressing strong self-interference signals from the transmitter using different methods in different parts of the system during simultaneous transmission and reception, including the propagation domain, analog domain, and digital domain. SIC is crucial for achieving simultaneous transmission and reception.

[0072] Ladder-wise grid structure (LWGS): This refers to a neural network structure used for digital self-interference cancellation in full-duplex transceivers. The network is trained using complex data and its connection between neurons in the input layer and the unique hidden layer is constructed based on a ladder-like topology.

[0073] Moving window grid structure (MWGS): This refers to a neural network structure used for digital self-interference cancellation in full-duplex transceivers. The network is trained using complex data, and its connection between neurons in the input layer and the unique hidden layer is constructed based on a moving window topology.

[0074] With the rapid development of the radio field, wireless frequency band resources are gradually becoming depleted. To address this challenge, many researchers have proposed the concept of simultaneous transmission and reception technology, also known as simultaneous full-duplex technology. This technology aims to transmit and receive electromagnetic wave signals simultaneously using the same medium resources, time, and frequency resources. Therefore, this technology can significantly improve the utilization rate of time and spectrum resources, but it also introduces a strong self-interference problem.

[0075] Over the past decade, simultaneous transmission and reception technology has made some progress. To suppress the influence of self-interference signals at the receiver as much as possible, three commonly used self-interference cancellation methods exist: propagation domain cancellation, analog domain cancellation, and digital domain cancellation. Generally, linear self-interference components can be modeled by weighted combination of different time delay terms, while nonlinear self-interference components can be characterized by polynomial models. However, polynomial models require a large number of model parameters and high computational complexity, especially when the nonlinear order is high. Therefore, neural networks have been developed to model nonlinear self-interference.

[0076] Most existing neural network architectures suffer from high computational complexity and poor self-interference suppression capabilities. While typical neural networks can suppress nonlinear self-interference to some extent when modeling it, the process of simulating nonlinearity does not conform to the actual generation patterns of nonlinear self-interference, requiring high computational complexity to ensure cancellation performance. Existing improved neural networks primarily reduce structural complexity by optimizing typical neural network structures or cascading different typical neural networks. While achieving some success, these efforts still fail to address the generation patterns of nonlinear self-interference, resulting in persistently high computational complexity.

[0077] In view of this, such as Figure 1 As shown, this embodiment of the invention provides a digital self-interference cancellation method based on an enhanced neural network structure, which may include, but is not limited to, steps S100 to S700:

[0078] Step S100: Perform complex value splitting and nonlinearization operations on the first stepped mesh structure to obtain the second stepped mesh structure, and perform complex value splitting and nonlinearization operations on the first sliding window mesh structure to obtain the second sliding window mesh structure.

[0079] Step S200: Obtain the first output of the dynamic phase layer according to the second stepped mesh structure or the second sliding window mesh structure;

[0080] Step S300: Construct a vector decomposition structure, and separate amplitude information and phase information based on the first output of the dynamic phase layer and the vector decomposition structure;

[0081] Step S400: Feed the amplitude information into the nonlinear hidden layer to obtain the second output of the nonlinear hidden layer;

[0082] Step S500: Perform a fusion operation on the phase information and the second output to obtain fused data;

[0083] Step S600: Feed the fused data into the fully connected output layer and output the simulation results;

[0084] Step S700: Based on the simulation results, perform self-interference cancellation operation on the received signal to obtain the target signal.

[0085] In steps S100 to S700 of some embodiments, the stepped mesh structure and sliding window mesh structure after complex value splitting and de-nonlinearization are used as dynamic phase layers. Then, the phase information and amplitude information are separated, and only the amplitude information is sent to the nonlinear hidden layer to characterize nonlinear self-interference. Finally, after recovering the phase information, the output is obtained through the fully connected layer.

[0086] In some embodiments, step S100 may include, but is not limited to, steps S101 to S104:

[0087] Step S101: The first complex value of the first stepped mesh structure is split into a first real part and a first imaginary part;

[0088] Step S102: Add a first neuron to the first hidden layer of the first stepped mesh structure;

[0089] Step S103: Combine the original second neuron of the first stepped grid structure with the first neuron, and replace the initial activation function of the first hidden layer with a linear activation function to obtain the second hidden layer;

[0090] Step S104: Based on the first real part, the first imaginary part, and the second hidden layer, the second stepped mesh structure is obtained.

[0091] In step S101 of some embodiments, the first complex value of the first stepped grid structure is split into a first real part and a first imaginary part, where the data format of both the first real part and the first imaginary part is real value, i.e., real number. Figure 2As shown, the first complex value x1(n) can be decomposed into the first real part. and the first imaginary part For example, the original complex number data is split into real and imaginary parts, and then combined to form real-valued input data, wherein the expression of the real-valued input data is:

[0092]

[0093] Where, x NN (n) represents the input to the neural network digital canceller; Indicates the first real part; Indicates the first imaginary part; [·] T This indicates the transpose operation.

[0094] In steps S102 to S103 of some embodiments, a new neuron is added to the output of each original hidden layer (i.e., the first hidden layer). The newly added hidden layer neuron and the original neurons are used to represent the real part and imaginary part of the hidden layer output data, respectively. Next, the activation function of the hidden layer in the first step-like grid structure is replaced with a linear activation function.

[0095] In some embodiments, prior to step S104, the output layer of the first stepped mesh structure, which is used to connect the vector decomposition structure in subsequent steps, is further included. Then, based on the first real part, the first imaginary part, and the second hidden layer, the following can be obtained: Figure 2 The enhanced stepped mesh structure shown is the second stepped mesh structure.

[0096] like Figure 2 As shown, the input data of the stepped grid structure is changed to a format where real and imaginary parts alternate, thus requiring an additional connection for the input at any given time step. Since the weights and biases in the neurons are all real values, an additional neuron is needed to describe the real and imaginary parts of the output data respectively. Furthermore, the nonlinear activation function in the hidden layers of the original stepped grid structure is replaced with a linear activation function, while the network connectivity remains unchanged. Figure 2 h1, h2, ..., h N This represents N neurons.

[0097] In some embodiments, step S100 may also include, but is not limited to, steps S111 to S114:

[0098] Step S111: The second complex value of the first sliding window mesh structure is split into a second real part and a second imaginary part;

[0099] Step S112: Add a third neuron to the third hidden layer of the first sliding window mesh structure;

[0100] Step S113: Combine the original fourth neuron of the first sliding window mesh structure with the third neuron, and replace the initial activation function of the third hidden layer with a linear activation function to obtain the fourth hidden layer;

[0101] Step S114: Based on the second real part, the second imaginary part, and the fourth hidden layer, the second sliding window mesh structure is obtained.

[0102] In step S111 of some embodiments, the second complex value of the first sliding window mesh structure is split into a second real part and a second imaginary part, where the data format of both the second real part and the second imaginary part is real value, i.e., real number. Figure 3 As shown, the second complex value x2(n) can be decomposed into the second real part. Second imaginary part For example, the original complex number data is split into real and imaginary parts, and then combined to form real-valued input data, wherein the expression of the real-valued input data is:

[0103]

[0104] Where, x NN (n) represents the input to the neural network digital canceller; Indicates the second real part; This represents the second imaginary part.

[0105] In steps S112 to S113 of some embodiments, a new neuron is added to the output of each original hidden layer (i.e., the third hidden layer). The newly added hidden layer neuron and the original neurons are used to represent the real and imaginary parts of the hidden layer output data, respectively. Next, the activation function of the hidden layer in the first sliding window grid structure is replaced with a linear activation function.

[0106] In some embodiments, prior to step S114, the output layer of the first sliding window mesh structure is deleted, which is used to connect the vector decomposition structure in subsequent steps. Then, based on the second real part, the second imaginary part, and the fourth hidden layer, the following can be obtained: Figure 3 The enhanced sliding window grid structure shown is the second sliding window grid structure.

[0107] like Figure 3 As shown, the input data of the sliding window mesh structure is changed to a format where real and imaginary parts alternate, thus requiring an additional connection for the input at a given time step. Since the weights and biases in the neurons are all real values, an additional neuron is needed to describe the real and imaginary parts of the output data respectively. Furthermore, the nonlinear activation function in the hidden layer of the original sliding window mesh structure is replaced with a linear activation function, while the network connection pattern remains unchanged. For ease of representation, optionally... Figure 3 The size of the sliding window in the sliding window grid structure is set to 2.

[0108] In step S200 of some embodiments, the second stepped mesh structure or the second sliding window mesh structure is used as the dynamic phase layer of the neural network. By applying the mesh design concepts of stepped mesh structures and sliding window mesh structures to the neural network, the network complexity can be reduced while ensuring digital self-interference cancellation performance. The output of the dynamic phase layer is then:

[0109] o ′ (n)=[o ′ 1(n),o ′ 2(n),…,o ′ 2G-1 (n),o ′ 2G (n)] T ;

[0110] Among them, o ′ (n) represents the first output of the dynamic phase layer; o ′ 1(n),o ′ 2(n),…,o ′ 2G-1 (n),o ′ 2G (n) represents the output of each neuron in the dynamic phase layer; G represents the number of neurons in the nonlinear hidden layer; n represents the discrete time step; [·] T This indicates the transpose operation.

[0111] In step S300 of some embodiments, a vector decomposition structure is designed and constructed. Based on the first output of the dynamic phase layer, amplitude information and phase information are separated through this vector decomposition structure. For example, the expression for the vector decomposition structure (VDS) is:

[0112]

[0113] Among them, s i (n) represents amplitude information; sinθ i Cosθ represents the phase of a sinusoidal wave. i G represents the cosine phase; G represents the number of neurons in the nonlinear hidden layer; n represents the discrete time step; o ′ 2i (n), o ′ 2i-1 (n) represents the output of the 2i-th and 2i-1-th neurons in the dynamic phase layer.

[0114] In step S400 of some embodiments, the amplitude information is fed into a nonlinear hidden layer with the activation function Tanh to obtain the second output of the nonlinear hidden layer. This step is crucial for simulating nonlinear self-interference, and the output conforms to the nonlinear operating rules. The amplitude information includes the magnitude and the square of the magnitude. For example, the input and output data of the nonlinear hidden layer are as follows:

[0115] x ′ NN (n)=[|s1(n)|,|s1(n)| 2 ,…,|s G (n)|,|s G (n)| 2 ] T ;

[0116] o ″ NN (n)=[o1 ″ (n),o2 ″ (n),…,o G ″ (n)] T ;

[0117] Where, x ′ NN (n) represents the input of the nonlinear hidden layer; s1(n),…,s G (n) represents amplitude information; [·] T Indicates the transpose operation; o ″ NN (n) represents the second output of the nonlinear hidden layer; o1 ″ (n),o2 ″ (n),…,o G ″ (n) represents the output of each neuron in the nonlinear hidden layer; NN represents a neural network.

[0118] In some embodiments, step S400 may include, but is not limited to, steps S401 to S402:

[0119] Step S401: Solve for the Taylor expansion of the Tanh function. For example, the Taylor expansion of the Tanh function is:

[0120]

[0121] Where tanh(x) represents the Tanh function; e x This represents a power function.

[0122] Step S402: Solve for the output expression of the nonlinear hidden layer. For example, the output expression of the nonlinear hidden layer is:

[0123]

[0124] Among them, w i (n)=[w i,1 (n),w i,2 (n),…,w i,G [(n)] represents the connection weight of the i-th neuron in the nonlinear hidden layer; b i (n) represents the bias of the i-th neuron in the nonlinear hidden layer.

[0125] In step S500 of some embodiments, a phase recover (PR) operation is designed to achieve phase recovery by fusing the amplitude information in the second output of the nonlinear hidden layer with the original phase information. For example, the expression of the fused data is:

[0126] x ″ NN (n)=[o1 ″ (n)sinθ1,o1 ″ (n)cosθ1,…,o G ″ (n)sinθ G ,o G ″ (n)cosθ G ] T ;

[0127] Where, x ″ NN (n) represents the fused data; o1 ″ (n),…,o G ″ (n) represents the output of each neuron in the nonlinear hidden layer; sinθ1,…,sinθ G Represents the phases of each sine wave; cosθ1,…,cosθ G Indicates the phase of each cosine; [·] T This indicates the transpose operation.

[0128] In step S600 of some embodiments, the fused data is fed into the fully connected output layer to obtain the real and imaginary parts of the self-interference estimate, i.e., the simulation result, which can be used to subtract the self-interference from the received signal.

[0129] In step S700 of some embodiments, based on the simulation results of a neural network canceller based on an enhanced stepped mesh structure or an enhanced sliding window mesh structure, a self-interference cancellation operation is performed on the received signal to obtain the remaining signal after cancellation, i.e., the target signal. For example, the expression for obtaining the target signal is:

[0130]

[0131] Where y(n) represents the target signal; y SI (n) represents the received signal; This indicates the simulation results.

[0132] In some embodiments, the digital self-interference cancellation method based on the enhanced neural network structure may further include: acquiring a full-duplex transmit / receive data set; splitting the full-duplex transmit / receive data set to obtain a training set and a test set; training a first neural network based on the second ladder-like grid structure according to the training set to obtain a second neural network; training a third neural network based on the second sliding window grid structure according to the training set to obtain a fourth neural network; and performing self-interference cancellation performance tests on the second neural network and the fourth neural network according to the test set to obtain test results.

[0133] For example, a publicly available dataset is obtained, which contains transmit and receive data of a 10MHz orthogonal frequency division multiplexing signal running on a full-duplex platform. This dataset is split into a training set and a test set. The training set is used to train neural networks based on an enhanced ladder grid structure and neural networks based on an enhanced sliding window grid structure, respectively. The test set is then used to test the self-interference cancellation performance of the trained enhanced neural networks.

[0134] like Figure 4 As shown, a neural network structure based on an enhanced ladder-like grid structure is described. The dynamic phase layer is a ladder-like grid structure after complex-valued decomposition and de-nonlinearization, with 2G neurons. Each pair of neurons forms a complex data input to the vector decomposition structure. The vector decomposition structure extracts amplitude and phase information respectively. The amplitude information includes the magnitude and its square, while the phase information is used for phase recovery of the input data to the output layer. The nonlinear hidden layer has G neurons, and the activation function is the Tanh function. After phase recovery of the output of the nonlinear hidden layer, the simulation result is output through a fully connected output layer.

[0135] Taking a neural network canceller based on an enhanced ladder-grid structure as an example, the implementation process of a digital self-interference cancellation method based on an enhanced neural network structure according to an embodiment of the present invention is described as follows:

[0136] Step 1: Perform complex-valued decomposition and nonlinearization on the stepped mesh structure;

[0137] Optionally, in this embodiment of the invention, the number of hidden layer neurons in the ladder-like grid structure after complex-valued splitting and de-nonlinearization can be 8, 12, or 16, corresponding to G values ​​of 4, 6, and 8, respectively.

[0138] Step 2: Use the complex-valued and de-nonlinearized stepped mesh structure as a dynamic phase layer.

[0139] Step 3: Design a vector decomposition structure to separate amplitude and phase information;

[0140] Optionally, the vector decomposition structure in this embodiment of the invention has G elements.

[0141] Step 4: Feed the amplitude information into the nonlinear hidden layer with the Tanh activation function, where the amplitude information includes the magnitude and the square of the magnitude.

[0142] Step 5: After phase recovery operation, fuse the amplitude information of the nonlinear hidden layer output with the phase information in step 2.

[0143] Step 6: Feed the fused data into the fully connected output layer to obtain the real and imaginary parts of the self-interference estimate, which are used to subtract the self-interference from the received signal.

[0144] Step 7: Split the dataset into a training set and a test set, and use the training set to train the enhanced LWGS network.

[0145] Step 8: Test the self-interference cancellation performance of the trained enhanced LWGS network using the test set.

[0146] like Figure 5 The diagram illustrates a neural network structure based on a sliding window mesh. The dynamic phase layer is a sliding window mesh structure after complex-valued decomposition and de-nonlinearization, with 2G neurons and a sliding window size of 2. Complex data is input to the vector decomposition structure in pairs of neurons. The vector decomposition structure extracts amplitude and phase information, including the magnitude and its square, while the phase information is used for phase recovery of the input data to the output layer. The nonlinear hidden layer has G neurons, and the activation function is the Tanh function. After phase recovery of the output of the nonlinear hidden layer, the simulation result is output through a fully connected output layer.

[0147] Taking a neural network canceller based on an enhanced sliding window mesh structure as an example, the implementation process of a digital self-interference cancellation method based on an enhanced neural network structure according to an embodiment of the present invention is described as follows:

[0148] Step 1: Perform complex-valued decomposition and nonlinearization on the sliding window mesh structure;

[0149] Step 2: Use the sliding window mesh structure after complex value splitting and de-nonlinearization as a dynamic phase layer.

[0150] Step 3: Design a vector decomposition structure to separate amplitude and phase information;

[0151] Optionally, the vector decomposition structure in this embodiment of the invention has G elements.

[0152] Step 4: Feed the amplitude information into the nonlinear hidden layer with the Tanh activation function, where the amplitude information includes the magnitude and the square of the magnitude.

[0153] Step 5: After phase recovery operation, fuse the amplitude information of the nonlinear hidden layer output with the phase information in step 2.

[0154] Step 6: Feed the fused data into the fully connected output layer to obtain the real and imaginary parts of the self-interference estimate, which are used to subtract the self-interference from the received signal.

[0155] Step 7: Split the dataset into a training set and a test set, and use the training set to train the enhanced LWGS network.

[0156] Step 8: Test the self-interference cancellation performance of the trained enhanced LWGS network using the test set.

[0157] refer to Figure 6 This describes a typical application scenario of an embodiment of the present invention, namely a nonlinear canceller. For example... Figure 6 As shown, x(n) is a digital baseband signal, x IQ (n) is the signal after the up mixer, x PA (n) is the transmitted signal after power amplification. After passing through a self-interference channel, low-noise amplifier (LNA), down-mixer, and analog-to-digital converter (ADC), it becomes y. SI (n), where, It is a linear self-interference time-domain response. This is based on simulation results of neural network cancellers using enhanced LWGS or enhanced MWGS. The remaining signal after cancellation is y(n). Therefore, the digital self-interference cancellation ratio can be defined to measure network performance:

[0158]

[0159] Wherein, γ represents dB This represents the digital self-interference cancellation ratio.

[0160] refer to Figure 7 Using the performance of a 5th-order nonlinear polynomial canceller as a reference, the results of neural network cancellers under different parameter settings are compared. Figure 7As shown, the self-interference cancellation performance of linear cancellers, typical polynomial cancellers, neural network cancellers based on enhanced ladder grid structure (ALWGS), and neural network cancellers based on enhanced sliding window grid structure (AMWGS) is described. In the figures, the numbers in the names represent the number of neurons; for example, the enhanced ladder grid structure (6) has G=6; the enhanced sliding window grid structure (3,5) has G=3, and the sliding window size is 5. Figure 7 The results show that the enhanced stepped grid structure and enhanced sliding window grid structure of the present invention can improve the digital self-interference cancellation performance.

[0161] This invention also provides a digital self-interference cancellation device based on an enhanced neural network structure, which can implement the above-mentioned digital self-interference cancellation method based on an enhanced neural network structure. The device includes:

[0162] The first module is used to perform complex value decomposition and nonlinearization operations on the first stepped mesh structure to obtain the second stepped mesh structure, and to perform complex value decomposition and nonlinearization operations on the first sliding window mesh structure to obtain the second sliding window mesh structure.

[0163] The second module is used to obtain the first output of the dynamic phase layer based on the second stepped mesh structure or the second sliding window mesh structure;

[0164] The third module is used to construct a vector decomposition structure, and to separate amplitude information and phase information based on the first output of the dynamic phase layer and the vector decomposition structure.

[0165] The fourth module is used to feed the amplitude information into the nonlinear hidden layer to obtain the second output of the nonlinear hidden layer;

[0166] The fifth module is used to perform a fusion operation on the phase information and the second output to obtain fused data;

[0167] The sixth module is used to feed the fused data into the fully connected output layer and output the simulation results;

[0168] The seventh module is used to perform self-interference cancellation operation on the received signal based on the simulation results to obtain the target signal.

[0169] It is understood that the content of the above method embodiments is applicable to the present device embodiments. The specific functions implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0170] This invention also provides an electronic device, which includes a processor and a memory. The memory stores a computer program, and when the processor executes the computer program, it implements the aforementioned digital self-interference cancellation method based on an enhanced neural network structure. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0171] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0172] refer to Figure 8 , Figure 8 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes:

[0173] The processor 801 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present invention.

[0174] The memory 802 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 802 can store the operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 802 and is called and executed by the processor 801 to implement the digital self-interference cancellation method based on an enhanced neural network structure according to the embodiments of the present invention.

[0175] The 803 input / output interface is used to implement information input and output.

[0176] The communication interface 804 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0177] Bus 805 transmits information between various components of the device (e.g., processor 801, memory 802, input / output interface 803, and communication interface 804);

[0178] The processor 801, memory 802, input / output interface 803, and communication interface 804 are connected to each other within the device via bus 805.

[0179] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned digital self-interference cancellation method based on an enhanced neural network structure.

[0180] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0181] This invention also provides a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium and execute the computer instructions, causing the computer device to perform the aforementioned digital self-interference cancellation method based on an enhanced neural network structure.

[0182] In summary, the digital self-interference cancellation method and apparatus based on an enhanced neural network structure according to embodiments of the present invention have the following advantages:

[0183] 1. The embodiments of the present invention combine the working law of nonlinearity to design the structure of the neural network, so that the neural network canceller has excellent performance in characterizing nonlinear self-interference.

[0184] 2. The embodiments of the present invention utilize the gridded design concepts of LWGS and MWGS, and apply them to the neural network to reduce network complexity while ensuring digital self-interference cancellation performance. The present invention can be widely applied in the fields of radar phased array and full-duplex communication technology.

[0185] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order shown in the operation diagrams. For example, depending on the functions / operations involved, two consecutively shown blocks may actually be executed substantially simultaneously, or the blocks may sometimes be executed in reverse order. Furthermore, the embodiments presented and described in the flowcharts of this invention are provided by way of example to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is altered and sub-operations described as part of a larger operation are executed independently.

[0186] Furthermore, although the invention has been described in the context of functional modules, it should be understood that, unless otherwise stated, one or more of the described functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the invention. Rather, given the properties, functions, and internal relationships of the various functional modules in the apparatus disclosed herein, the actual implementation of the module will be understood within the scope of conventional skill of an engineer. Therefore, those skilled in the art can implement the invention as set forth in the claims using ordinary techniques without excessive experimentation. It is also understood that the specific concepts disclosed are merely illustrative and not intended to limit the scope of the invention, which is determined by the full scope of the appended claims and their equivalents.

[0187] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0188] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0189] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0190] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0191] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0192] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

[0193] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of the present invention.

Claims

1. A digital self-interference cancellation method based on an enhanced neural network structure, characterized in that, Includes the following steps: The first step mesh structure is subjected to complex value decomposition and nonlinearity removal operations to obtain the second step mesh structure. The first sliding window mesh structure is subjected to complex value decomposition and nonlinearity removal operations to obtain the second sliding window mesh structure. Based on the second stepped mesh structure or the second sliding window mesh structure, obtain the first output of the dynamic phase layer; A vector decomposition structure is constructed, and amplitude information and phase information are separated based on the first output of the dynamic phase layer and the vector decomposition structure. The amplitude information is fed into the nonlinear hidden layer to obtain the second output of the nonlinear hidden layer; The phase information and the second output are fused to obtain fused data; The fused data is fed into a fully connected output layer to output simulation results; Based on the simulation results, a self-interference cancellation operation is performed on the received signal to obtain the target signal; The step of performing complex value decomposition and nonlinearization operations on the first-order mesh structure to obtain the second-order mesh structure includes the following steps: The first complex value of the first stepped mesh structure is split into a first real part and a first imaginary part; A first neuron is added to the first hidden layer of the first stepped grid structure; The original second neuron of the first step grid structure is combined with the first neuron, and the initial activation function of the first hidden layer is replaced with a linear activation function to obtain the second hidden layer; The second stepped mesh structure is obtained based on the first real part, the first imaginary part, and the second hidden layer; The process of performing complex value decomposition and nonlinearity removal operations on the first sliding window mesh structure to obtain the second sliding window mesh structure includes the following steps: The second complex value of the first sliding window mesh structure is split into a second real part and a second imaginary part; A third neuron is added to the third hidden layer of the first sliding window mesh structure; The original fourth neuron of the first sliding window mesh structure is combined with the third neuron, and the initial activation function of the third hidden layer is replaced with a linear activation function to obtain the fourth hidden layer. The second sliding window mesh structure is obtained based on the second real part, the second imaginary part, and the fourth hidden layer.

2. The digital self-interference cancellation method based on an enhanced neural network structure according to claim 1, characterized in that, It also includes the following steps: Obtain the full-duplex transmit and receive data set; The full-duplex transmit and receive data set is split to obtain a training set and a test set; Based on the training set, the first neural network based on the second ladder grid structure is trained to obtain the second neural network; Based on the training set, the third neural network based on the second sliding window mesh structure is trained to obtain the fourth neural network; Based on the test set, the self-interference cancellation performance of the second neural network and the fourth neural network is tested, and the test results are obtained.

3. The digital self-interference cancellation method based on an enhanced neural network structure according to claim 1, characterized in that, The formula used to obtain the first output of the dynamic phase layer based on the second stepped mesh structure or the second sliding window mesh structure includes: ； in, This indicates the first output of the dynamic phase layer; This represents the output of each neuron in the dynamic phase layer; This represents the number of neurons in a non-linear hidden layer; Represents discrete time points; This indicates the transpose operation.

4. The digital self-interference cancellation method based on an enhanced neural network structure according to claim 1, characterized in that, The constructed vector decomposition structure, based on the first output of the dynamic phase layer and the vector decomposition structure, separates amplitude information and phase information. The formulas used include: ； ； ； in, Indicates amplitude information; Indicates the phase of a sine wave; Indicates the cosine phase; This represents the number of neurons in a non-linear hidden layer; Represents discrete time points; , Represents the dynamic phase layer. The and the first The output of each neuron.

5. The digital self-interference cancellation method based on an enhanced neural network structure according to claim 1, characterized in that, The formula used to feed the amplitude information into the nonlinear hidden layer to obtain the second output of the nonlinear hidden layer includes: ； ； in, This represents the input to the nonlinear hidden layer; Indicates amplitude information; Indicates the transpose operation; This represents the second output of the non-linear hidden layer; This represents the output of each neuron in the nonlinear hidden layer.

6. The digital self-interference cancellation method based on an enhanced neural network structure according to claim 1, characterized in that, The formula used to perform the fusion operation on the phase information and the second output to obtain fused data includes: ； in, This indicates fused data; This represents the output of each neuron in the nonlinear hidden layer; Indicates the phases of each sine wave; Indicates the phase of each cosine; This indicates the transpose operation.

7. The digital self-interference cancellation method based on an enhanced neural network structure according to claim 1, characterized in that, The step of performing self-interference cancellation on the received signal based on the simulation results to obtain the target signal uses the following formulas: ； in, Indicates the target signal; Indicates the received signal; This indicates the simulation results.

8. A digital self-interference cancellation device based on an enhanced neural network structure, characterized in that, include: The first module is used to perform complex value decomposition and nonlinearization operations on the first stepped mesh structure to obtain the second stepped mesh structure, and to perform complex value decomposition and nonlinearization operations on the first sliding window mesh structure to obtain the second sliding window mesh structure. The second module is used to obtain the first output of the dynamic phase layer based on the second stepped mesh structure or the second sliding window mesh structure; The third module is used to construct a vector decomposition structure, and to separate amplitude information and phase information based on the first output of the dynamic phase layer and the vector decomposition structure. The fourth module is used to feed the amplitude information into the nonlinear hidden layer to obtain the second output of the nonlinear hidden layer; The fifth module is used to perform a fusion operation on the phase information and the second output to obtain fused data; The sixth module is used to feed the fused data into the fully connected output layer and output the simulation results; The seventh module is used to perform self-interference cancellation operation on the received signal based on the simulation results to obtain the target signal; Specifically, the first module is used for: The first complex value of the first stepped mesh structure is split into a first real part and a first imaginary part; A first neuron is added to the first hidden layer of the first stepped grid structure; The original second neuron of the first step grid structure is combined with the first neuron, and the initial activation function of the first hidden layer is replaced with a linear activation function to obtain the second hidden layer; The second stepped mesh structure is obtained based on the first real part, the first imaginary part, and the second hidden layer; The second complex value of the first sliding window mesh structure is split into a second real part and a second imaginary part; A third neuron is added to the third hidden layer of the first sliding window mesh structure; The original fourth neuron of the first sliding window mesh structure is combined with the third neuron, and the initial activation function of the third hidden layer is replaced with a linear activation function to obtain the fourth hidden layer. The second sliding window mesh structure is obtained based on the second real part, the second imaginary part, and the fourth hidden layer.

Images