Coherent signal processing method, apparatus and device, and readable storage medium

Abstract

A coherent signal processing method, apparatus and device, and a readable storage medium, relating to the technical field of communications, comprising: inputting a coherent complex valued signal to be processed to a target neuron model, controlling a complex convolution layer in the target neuron model to perform convolution operation on the coherent complex valued signal to be processed and neurons in the complex convolution layer, obtaining a first output signal; performing nonlinear transformation on the first output signal by a plurality of activation layers in the target neuron model to obtain a second output signal; flattening the second output signal by a plurality of flattening layers in the target neuron model to obtain a target one-dimensional vector; and a plurality of full connection layers in the target neuron model perform feature mapping on the target one-dimensional vector to output a target coherent complex value signal. According to the invention, the orthogonality between the I path and the Q path is fully considered, and the quality of the output signal can be effectively improved.

Description

Technical Field

[0001] The present application relates to the field of communication technology, and in particular to a coherent signal processing method, device, equipment and readable storage medium. Background Art

[0002] In a coherent transmission system, the received coherent signal is usually expressed in complex form, that is, it includes a real part (In-phase, I path) and an imaginary part (Quadrature, Q path). These two parts represent the components of the signal in two orthogonal phases, and the I path and the Q path jointly carry the complete information of the signal (including amplitude and phase), that is, there is an inherent phase and amplitude relationship between the two.

[0003] There are many types of neuron models that are used in coherent transmission systems. Figure 1 As shown in the figure, when a neuron-like model is used to process data in a traditional coherent transmission system, the I and Q paths of the coherent signal are usually separated and then independently input into the neuron-like model to obtain two real output paths, which are then recombined into complex signals for output. It can be seen that the above operation ignores the correlation between the I and Q paths in the coherent signal, so that the model cannot fully learn the correlation between the I and Q paths in the coherent signal, so that the nonlinear distortion in the coherent system cannot be effectively processed and the channel estimation results are affected, resulting in a decrease in the quality of the output signal. Summary of the invention

[0004] The present application provides a coherent signal processing method, device, equipment and readable storage medium, which can solve the technical problem in the prior art that the quality of the output signal is reduced due to failure to consider the correlation between the I path and the Q path.

[0005] In a first aspect, an embodiment of the present application provides a coherent signal processing method, which is applied to a target class neuron model, wherein the target class neuron model includes a complex convolution layer, a complex activation layer, a complex flattening layer, and a complex fully connected layer, and the method includes:

[0006] Inputting the coherent complex-valued signal to be processed into the target class neuron model;

[0007] Control the target class neuron model to output a target coherent complex-valued signal based on the coherent complex-valued signal to be processed; wherein the complex convolution layer is used to perform a convolution operation on the coherent complex-valued signal to be processed and the neurons in the complex convolution layer to obtain a first output signal; the complex activation layer is used to perform a nonlinear transformation on the first output signal to obtain a second output signal; the complex flattening layer is used to flatten the second output signal to obtain a target one-dimensional vector; the complex fully connected layer is used to perform feature mapping on the target one-dimensional vector to output a target coherent complex-valued signal.

[0008] In combination with the first aspect, in one implementation, the complex convolution layer is used to perform a convolution operation on the coherent complex-valued signal to be processed and the neurons in the complex convolution layer to obtain a first output signal, including:

[0009] The complex convolution layer performs a convolution operation on the coherent complex-valued signal to be processed and the neurons in the complex convolution layer based on a first calculation formula to output a first real signal;

[0010] The complex convolution layer performs a convolution operation on the coherent complex-valued signal to be processed and the neurons in the complex convolution layer based on a second calculation formula to output a first imaginary signal;

[0011] The complex convolution layer obtains a first output signal based on a third calculation formula, the first real signal and the first imaginary signal;

[0012] Among them, the first calculation formula is:

[0013]

[0014] The second calculation formula is:

[0015]

[0016] The third calculation formula is:

[0017]

[0018] In the formula, S i,k represents the first output signal at position i and output channel k, j represents the imaginary unit, represents the first real signal at position i and output channel k, represents the first imaginary signal at position i and output channel k, M represents the length of the neuron, represents the real part signal in the coherent complex valued signal to be processed at position i+m, represents the imaginary signal in the coherent complex-valued signal to be processed at position i+m, represents the real neuron weight at position m and output channel k, represents the imaginary neuron weight at position m and output channel k, represents the real bias at output channel k, Represents the imaginary bias at output channel k.

[0019] In combination with the first aspect, in one implementation, the complex activation layer is used to perform a nonlinear transformation on the first output signal to obtain a second output signal, including:

[0020] The complex activation layer performs a nonlinear transformation on the first real signal in the first output signal based on a fourth calculation formula to output a second real signal;

[0021] The complex activation layer performs a nonlinear transformation on the first imaginary signal in the first output signal based on a fifth calculation formula to output a second imaginary signal;

[0022] The complex activation layer obtains a second output signal based on the sixth calculation formula, the second real signal and the second imaginary signal;

[0023] Among them, the fourth calculation formula is:

[0024]

[0025] The fifth calculation formula is:

[0026]

[0027] The sixth calculation formula is:

[0028]

[0029] In the formula, y i,k represents the second output signal at position i and output channel k, represents the second real signal at position i and output channel k, Represents the second imaginary signal at position i and output channel k.

[0030] In combination with the first aspect, in one embodiment, the target class neuron model also includes a complex normalization layer, which is used to normalize the second output signal and input the normalized second output signal to the complex flattening layer.

[0031] In a second aspect, an embodiment of the present application provides a coherent signal processing device, the coherent signal processing device comprising:

[0032] An input module, which is used to input the coherent complex-valued signal to be processed into a target-type neuron model, wherein the target-type neuron model includes a complex convolution layer, a complex activation layer, a complex flattening layer, and a complex fully connected layer;

[0033] A control module, which is used to control the target class neuron model to output a target coherent complex-valued signal based on the coherent complex-valued signal to be processed; wherein the complex convolution layer is used to perform a convolution operation on the coherent complex-valued signal to be processed and the neurons in the complex convolution layer to obtain a first output signal; the complex activation layer is used to perform a nonlinear transformation on the first output signal to obtain a second output signal; the complex flattening layer is used to flatten the second output signal to obtain a target one-dimensional vector; the complex fully connected layer is used to perform feature mapping on the target one-dimensional vector to output a target coherent complex-valued signal.

[0034] In combination with the second aspect, in one implementation, the complex convolution layer is used to perform a convolution operation on the coherent complex-valued signal to be processed and the neurons in the complex convolution layer based on a first calculation formula to output a first real signal;

[0035] The complex convolution layer is used to perform a convolution operation on the coherent complex-valued signal to be processed and the neurons in the complex convolution layer based on a second calculation formula to output a first imaginary signal;

[0036] The complex convolution layer is used to obtain a first output signal based on a third calculation formula, a first real signal and a first imaginary signal;

[0037] Among them, the first calculation formula is:

[0038]

[0039] The second calculation formula is:

[0040]

[0041] The third calculation formula is:

[0042]

[0043] In the formula, S i,k represents the first output signal at position i and output channel k, j represents the imaginary unit, represents the first real signal at position i and output channel k, represents the first imaginary signal at position i and output channel k, M represents the length of the neuron, represents the real part signal in the coherent complex valued signal to be processed at position i+m, represents the imaginary signal in the coherent complex-valued signal to be processed at position i+m, w m R ,k represents the real neuron weight at position m and output channel k, represents the imaginary neuron weight at position m and output channel k, represents the real bias at output channel k, Represents the imaginary bias at output channel k.

[0044] In combination with the second aspect, in one implementation, the complex activation layer is used to perform a nonlinear transformation on a first real signal in the first output signal based on a fourth calculation formula to output a second real signal; the complex activation layer is used to perform a nonlinear transformation on a first imaginary signal in the first output signal based on a fifth calculation formula to output a second imaginary signal; the complex activation layer is used to obtain a second output signal based on a sixth calculation formula, the second real signal, and the second imaginary signal;

[0045] Among them, the fourth calculation formula is:

[0046]

[0047] The fifth calculation formula is:

[0048]

[0049] The sixth calculation formula is:

[0050]

[0051] In the formula, y i,k represents the second output signal at position i and output channel k, represents the second real signal at position i and output channel k, Represents the second imaginary signal at position i and output channel k.

[0052] In combination with the second aspect, in one embodiment, the target class neuron model also includes a complex normalization layer, which is used to normalize the second output signal and input the normalized second output signal to the complex flattening layer.

[0053] In a third aspect, an embodiment of the present application provides a coherent signal processing device, which includes a processor, a memory, and a coherent signal processing program stored in the memory and executable by the processor, wherein when the coherent signal processing program is executed by the processor, the steps of the coherent signal processing method as described above are implemented.

[0054] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, on which a coherent signal processing program is stored, wherein when the coherent signal processing program is executed by a processor, the steps of the coherent signal processing method as described above are implemented.

[0055] The beneficial effects brought by the technical solution provided in the embodiments of the present application include:

[0056] The model input, the output of each layer in the model and the final output are all complex values ​​through the principle of complex number operations, that is, the coherent complex-valued signal to be processed is input into the target neuron model, so that the complex convolution layer in the target neuron model performs convolution operation on the coherent complex-valued signal to be processed and the neurons in the complex convolution layer, the complex activation layer performs nonlinear transformation on the signal output after the convolution operation, the complex flattening layer flattens the signal output by the complex activation layer, and the complex fully connected layer performs feature mapping on the flattened target one-dimensional vector and outputs the target coherent complex-valued signal. It fully considers the orthogonality between the I path and the Q path, so as to better solve a series of similar problems such as nonlinear distortion and channel estimation in the coherent transmission system, thereby effectively improving the quality of the output signal. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 Schematic diagram of the coherent signal processing flow in the prior art

[0058] Figure 2 This is a flow chart of an embodiment of a coherent signal processing method of the present application;

[0059] Figure 3 This is a schematic diagram of the overall processing flow of the coherent signal involved in the embodiment of the present application;

[0060] Figure 4 A schematic diagram of a specific processing flow of coherent signals involved in the embodiment of the present application;

[0061] Figure 5 Schematic diagram of the hardware structure of the coherent signal processing device involved in the embodiment of the present application. DETAILED DESCRIPTION

[0062] 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.

[0063] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0064] In a first aspect, an embodiment of the present application provides a coherent signal processing method.

[0065] In one embodiment, referring to Figure 2 , Figure 2 The flowchart of the embodiment of the coherent signal processing method of the present application is shown in FIG. The coherent signal processing method is applied to a target class neuron model, and the target class neuron model includes a complex convolution layer, a complex activation layer, a complex flattening layer, and a complex fully connected layer; Figure 2 As shown, the coherent signal processing method includes:

[0066] Step S10: inputting the coherent complex-valued signal to be processed into the target neuron model.

[0067] Exemplarily, in this embodiment, the target class neuron model is composed of a complex convolution layer, a complex activation layer, a complex flattening layer, and a complex fully connected layer, etc.; wherein the complex convolution layer is used to perform convolution operations, the complex activation layer is used to introduce nonlinear transformations, the complex flattening layer is used to perform flattening operations, and the complex fully connected layer is used to perform feature mapping to match different output dimension requirements. For details, see Figure 3 As shown, this embodiment uses the coherent complex-valued signal to be processed as the input of the target neuron model, so that the target neuron model performs correlation processing on the coherent complex-valued signal to be processed and then outputs the generated target complex-valued signal.

[0068] Step S20: Control the target class neuron model to output a target coherent complex valued signal based on the coherent complex valued signal to be processed; wherein the complex convolution layer is used to perform a convolution operation on the coherent complex valued signal to be processed and the neurons in the complex convolution layer to obtain a first output signal; the complex activation layer is used to perform a nonlinear transformation on the first output signal to obtain a second output signal; the complex flattening layer is used to flatten the second output signal to obtain a target one-dimensional vector; the complex fully connected layer is used to perform feature mapping on the target one-dimensional vector to output a target coherent complex valued signal. Furthermore, the target class neuron model also includes a complex normalization layer, which is used to normalize the second output signal and input the normalized second output signal to the complex flattening layer.

[0069] For example, see Figure 4 As shown in a, when the coherent complex-valued signal to be processed enters the complex convolution layer, the complex convolution layer will be based on Figure 4 The complex multiplication principle in b performs convolution operation on the coherent complex valued signal to be processed and the neurons of the complex convolution layer, so that the real part of the product obtained by the complex multiplication is the product of the real parts of the two multipliers minus the product of the imaginary parts of the two multipliers, and the imaginary part is the sum of the real and imaginary parts of the two multipliers. In order to enable the model to cope with more complex transmission environments, this embodiment will introduce nonlinear transformation through the complex activation layer; specifically, the signal output from the complex convolution layer will be nonlinearly transformed through the complex activation layer, that is, the complex activation layer will be based on Figure 4 The complex addition operation principle in c performs a nonlinear transformation on the signal output by the complex convolutional layer, that is, the real part of the sum obtained by adding two complex numbers is the result of adding the real parts of the addends, and the imaginary part of the sum is the result of adding the imaginary parts of the addends.

[0070] In order to cooperate with the gradient descent algorithm of the neuron-like model itself and increase the generalization ability of the model to accelerate the subsequent model fitting speed, a complex normalization layer is added to the target neuron-like model of this embodiment to normalize the signal after the complex activation layer is activated by the complex normalization layer. It should be noted that the operation principle of the complex normalization layer is still Figure 4 Complex number addition operations in c.

[0071] Then, the signal normalized by the complex normalization layer is flattened by the complex flattening layer to flatten it into a one-dimensional vector (i.e., the target one-dimensional vector); due to different application scenarios, the required output dimensions of the neuron-like models are also different. Therefore, in order to meet the model output dimension requirements corresponding to different applications, a complex fully connected layer is added to the end of the target neuron-like model, and based on Figure 4 The complex multiplication principle in a operates on the one-dimensional vector output by the complex flattening layer to output the final target coherent complex-valued signal. It can be seen that this embodiment directly inputs the coherent complex-valued signal into the model as a whole, which fully considers the correlation between the I path and the Q path, so as to make it easier for the model to learn signal characteristics, thereby enabling the model to better cope with different problems existing in coherent transmission.

[0072] Furthermore, in one embodiment, the complex convolution layer is used to perform a convolution operation on the coherent complex-valued signal to be processed and the neurons in the complex convolution layer to obtain a first output signal, including:

[0073] The complex convolution layer performs a convolution operation on the coherent complex-valued signal to be processed and the neurons in the complex convolution layer based on a first calculation formula to output a first real signal;

[0074] The complex convolution layer performs a convolution operation on the coherent complex-valued signal to be processed and the neurons in the complex convolution layer based on a second calculation formula to output a first imaginary signal;

[0075] The complex convolution layer obtains a first output signal based on a third calculation formula, the first real signal and the first imaginary signal;

[0076] Among them, the first calculation formula is:

[0077]

[0078] The second calculation formula is:

[0079]

[0080] The third calculation formula is:

[0081]

[0082] In the formula, S i,k represents the first output signal at position i and output channel k, j represents the imaginary unit, represents the first real signal at position i and output channel k, represents the first imaginary signal at position i and output channel k, M represents the length of the neuron, represents the real part signal in the coherent complex valued signal to be processed at position i+m, represents the imaginary signal in the coherent complex-valued signal to be processed at position i+m, represents the real neuron weight at position m and output channel k, represents the imaginary neuron weight at position m and output channel k, represents the real bias at output channel k, Represents the imaginary bias at output channel k.

[0083] Exemplarily, in this embodiment, the complex convolution layer performs a convolution operation on the input signal x (i.e., the coherent complex-valued signal to be processed) and the neurons of the complex convolution layer to obtain an output signal of the complex convolution layer:

[0084]

[0085]

[0086]

[0087] In the formula, S i,k represents the signal at position i and output channel k of the complex convolutional layer output, j represents the imaginary unit, Represents the real signal at position i and output channel k of the complex convolutional layer output, represents the imaginary signal at position i and output channel k of the complex convolutional layer output, M represents the length of the neuron, Represents the real part of the input signal x corresponding to the position i+m, Represents the imaginary signal in the input signal x corresponding to the position i+m, represents the real neuron weight at position m and output channel k, represents the imaginary neuron weight at position m and output channel k, represents the real bias at output channel k, Represents the imaginary bias at output channel k.

[0088] Furthermore, in one embodiment, the complex activation layer is used to perform a nonlinear transformation on the first output signal to obtain a second output signal, including:

[0089] The complex activation layer performs a nonlinear transformation on the first real signal in the first output signal based on a fourth calculation formula to output a second real signal;

[0090] The complex activation layer performs a nonlinear transformation on the first imaginary signal in the first output signal based on a fifth calculation formula to output a second imaginary signal;

[0091] The complex activation layer obtains a second output signal based on the sixth calculation formula, the second real signal and the second imaginary signal;

[0092] Among them, the fourth calculation formula is:

[0093]

[0094] The fifth calculation formula is:

[0095]

[0096] The sixth calculation formula is:

[0097]

[0098] In the formula, y i,k represents the second output signal at position i and output channel k, represents the second real signal at position i and output channel k, Represents the second imaginary signal at position i and output channel k.

[0099] For example, it should be noted that the activation function of the complex activation layer can be sigmoid, or other activation functions such as ReLu that can be applied to the neuron-like model; in this embodiment, sigmoid can be preferably used as the activation function of the complex activation layer, then the real signal of the complex convolution layer output by the complex activation layer is and the imaginary signal The output signal of the complex activation layer can be obtained by performing the following operations:

[0100]

[0101]

[0102]

[0103] In the formula, y i,krepresents the signal at position i and output channel k of the complex activation layer output, represents the real signal at position i and output channel k of the complex activation layer output, Represents the imaginary signal at position i and output channel k of the complex activation layer output.

[0104] Next, the data processed by the complex activation layer is sent to the complex normalization layer for the following normalization operation to obtain the output signal of the complex normalization layer:

[0105]

[0106]

[0107]

[0108] In the formula, represents the signal at position i and output channel k of the complex normalization layer output, represents the real signal at position i and output channel k of the complex normalization layer output, Represents the imaginary signal at position i and output channel k of the complex normalization layer output; ∈ represents a very small number, which is used to prevent the denominator from being zero. The specific value of ∈ can be determined according to actual needs and is not limited here; and Respectively The mean and variance of and Respectively The mean and variance of , and the specific calculation formula is as follows:

[0109]

[0110]

[0111]

[0112]

[0113] Where N represents the length of each batch of input model symbols.

[0114] Then the normalized data output by the complex normalization layer is flattened as follows:

[0115]

[0116]

[0117]

[0118] V=VR +V I j

[0119] Among them, L represents the length of the input data, K represents the number of neurons in the complex convolutional layer, that is,

[0120] is the number of output channels; V is a one-dimensional vector after flattening, V R express After normalization, V I express The normalized result.

[0121] Finally, the one-dimensional vector V is passed through a complex fully connected layer to output the required feature dimension. This embodiment takes the output of one feature based on the one-dimensional vector V as an example. Of course, other feature dimensions can also be used in other applications:

[0122]

[0123]

[0124]

[0125] z=z R +z I i

[0126] Where W represents a weight matrix with a dimension of 1×(L×K), ω L×K represents the weight corresponding to the L×Kth element, represents the real weight corresponding to the i-th element, represents the imaginary weight corresponding to the i-th element, v L×K represents the vector value corresponding to the L×Kth element, Represents the real vector value corresponding to the i-th element, represents the imaginary vector value corresponding to the i-th element, b R represents the real bias at the output channel, b I represents the imaginary bias at the output channel, z R represents the real signal of the complex fully connected layer output, z I represents the imaginary signal output by the complex fully connected layer, and z represents the signal output by the complex fully connected layer.

[0127] It should be noted that this embodiment only shows the calculation process of the target neuron model once. In practical applications, the above operation can be cycled n times according to specific requirements to meet the transmission indicators of the coherent transmission system.

[0128] In a second aspect, an embodiment of the present application also provides a coherent signal processing device.

[0129] In one embodiment, the coherent signal processing device includes:

[0130] An input module, which is used to input the coherent complex-valued signal to be processed into a target-type neuron model, wherein the target-type neuron model includes a complex convolution layer, a complex activation layer, a complex flattening layer, and a complex fully connected layer;

[0131] A control module, which is used to control the target class neuron model to output a target coherent complex-valued signal based on the coherent complex-valued signal to be processed; wherein the complex convolution layer is used to perform a convolution operation on the coherent complex-valued signal to be processed and the neurons in the complex convolution layer to obtain a first output signal; the complex activation layer is used to perform a nonlinear transformation on the first output signal to obtain a second output signal; the complex flattening layer is used to flatten the second output signal to obtain a target one-dimensional vector; the complex fully connected layer is used to perform feature mapping on the target one-dimensional vector to output a target coherent complex-valued signal.

[0132] Further, in one embodiment, the complex convolution layer is used to perform a convolution operation on the coherent complex-valued signal to be processed and the neurons in the complex convolution layer based on a first calculation formula to output a first real signal; the complex convolution layer is used to perform a convolution operation on the coherent complex-valued signal to be processed and the neurons in the complex convolution layer based on a second calculation formula to output a first imaginary signal; the complex convolution layer is used to obtain a first output signal based on a third calculation formula, the first real signal and the first imaginary signal;

[0133] Among them, the first calculation formula is:

[0134]

[0135] The second calculation formula is:

[0136]

[0137] The third calculation formula is:

[0138]

[0139] In the formula, S i,k represents the first output signal at position i and output channel k, j represents the imaginary unit, represents the first real signal at position i and output channel k, represents the first imaginary signal at position i and output channel k, M represents the length of the neuron, represents the real part signal in the coherent complex valued signal to be processed at position i+m, represents the imaginary signal in the coherent complex-valued signal to be processed at position i+m, represents the real neuron weight at position m and output channel k, represents the imaginary neuron weight at position m and output channel k, represents the real bias at output channel k, Represents the imaginary bias at output channel k.

[0140] Further, in one embodiment, the complex activation layer is used to perform a nonlinear transformation on the first real signal in the first output signal based on a fourth calculation formula to output a second real signal; the complex activation layer is used to perform a nonlinear transformation on the first imaginary signal in the first output signal based on a fifth calculation formula to output a second imaginary signal; the complex activation layer is used to obtain a second output signal based on a sixth calculation formula, the second real signal and the second imaginary signal;

[0141] Among them, the fourth calculation formula is:

[0142]

[0143] The fifth calculation formula is:

[0144]

[0145] The sixth calculation formula is:

[0146]

[0147] In the formula, y i,k represents the second output signal at position i and output channel k, represents the second real signal at position i and output channel k, Represents the second imaginary signal at position i and output channel k.

[0148] Furthermore, in one embodiment, the target class neuron model also includes a complex normalization layer, and the complex normalization layer is used to normalize the second output signal and input the normalized second output signal to the complex flattening layer.

[0149] Among them, the functional implementation of each module in the above-mentioned coherent signal processing device corresponds to each step in the above-mentioned coherent signal processing method embodiment, and its functions and implementation processes are no longer repeated here.

[0150] In a third aspect, an embodiment of the present application provides a coherent signal processing device, which may be a personal computer (PC), a laptop computer, a server, or other device with a data processing function.

[0151] Reference Figure 5 , Figure 5 The hardware structure diagram of the coherent signal processing device involved in the embodiment of the present application is shown in FIG. In the embodiment of the present application, the coherent signal processing device may include a processor, a memory, a communication interface, and a communication bus.

[0152] The communication bus may be of any type and is used to interconnect the processor, the memory, and the communication interface.

[0153] The communication interface includes input / output (I / O) interfaces, physical interfaces, and logical interfaces, which are used to interconnect the devices inside the coherent signal processing device, and interfaces used to interconnect the coherent signal processing device with other devices (such as other computing devices or user devices). The physical interface can be an Ethernet interface, a fiber optic interface, an ATM interface, etc.; the user device can be a display, a keyboard, etc.

[0154] The memory can be various types of storage media, such as random access memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), flash memory, optical storage, hard disk, programmable ROM (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), etc.

[0155] The processor may be a general-purpose processor, and the general-purpose processor may call the coherent signal processing program stored in the memory and execute the coherent signal processing method provided in the embodiment of the present application. For example, the general-purpose processor may be a central processing unit (CPU). The method executed when the coherent signal processing program is called may refer to the various embodiments of the coherent signal processing method of the present application, and will not be repeated here.

[0156] Those skilled in the art will understand that Figure 5 The hardware structure shown in the figure does not constitute a limitation on the present application, and may include more or less components than shown in the figure, or combine certain components, or arrange the components differently.

[0157] In a fourth aspect, an embodiment of the present application also provides a computer-readable storage medium.

[0158] The readable storage medium of the present application stores a coherent signal processing program, wherein when the coherent signal processing program is executed by a processor, the steps of the coherent signal processing method as described above are implemented.

[0159] Among them, the method implemented when the coherent signal processing program is executed can refer to the various embodiments of the coherent signal processing method of the present application, and will not be repeated here.

[0160] It should be noted that the serial numbers of the above-mentioned embodiments of the present application are only for description and do not represent the advantages or disadvantages of the embodiments.

[0161] The terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned drawings are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally includes steps or units that are not listed, or optionally includes other steps or units inherent to these processes, methods, products or devices. The terms "first", "second" and "third" are used to distinguish different objects, etc., and do not represent a sequence, nor do they limit "first", "second" and "third" to different types.

[0162] In the description of the embodiments of the present application, "exemplary", "for example" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary", "for example" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary", "for example" or "for example" is intended to present related concepts in a specific way.

[0163] In the description of the embodiments of the present application, unless otherwise specified, “ / ” means or, for example, A / B can mean A or B; the “and / or” in the text is merely a description of the association relationship of associated objects, indicating that three relationships may exist, for example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, in the description of the embodiments of the present application, “multiple” refers to two or more than two.

[0164] In some processes described in the embodiments of the present application, multiple operations or steps that appear in a specific order are included, but it should be understood that these operations or steps may not be executed in the order in which they appear in the embodiments of the present application or in parallel, and the sequence number of the operation is only used to distinguish the different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed in sequence or in parallel, and these operations or steps may be combined.

[0165] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, disk, CD) as described above, and includes a number of instructions for a terminal device to execute the methods described in each embodiment of the present application.

[0166] The above are only preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A coherent signal processing method, characterized in that: The coherent signal processing method is applied to a target class neuron model, wherein the target class neuron model includes a complex convolution layer, a complex activation layer, a complex flattening layer and a complex fully connected layer, and the method includes: Inputting the coherent complex-valued signal to be processed into the target class neuron model; Control the target class neuron model to output a target coherent complex-valued signal based on the coherent complex-valued signal to be processed; wherein the complex convolution layer is used to perform a convolution operation on the coherent complex-valued signal to be processed and the neurons in the complex convolution layer to obtain a first output signal; the complex activation layer is used to perform a nonlinear transformation on the first output signal to obtain a second output signal; the complex flattening layer is used to flatten the second output signal to obtain a target one-dimensional vector; the complex fully connected layer is used to perform feature mapping on the target one-dimensional vector to output a target coherent complex-valued signal.

2. The coherent signal processing method according to claim 1, characterized in that: The complex convolution layer is used to perform a convolution operation on the coherent complex-valued signal to be processed and the neurons in the complex convolution layer to obtain a first output signal, including: The complex convolution layer performs a convolution operation on the coherent complex-valued signal to be processed and the neurons in the complex convolution layer based on a first calculation formula to output a first real signal; The complex convolution layer performs a convolution operation on the coherent complex-valued signal to be processed and the neurons in the complex convolution layer based on a second calculation formula to output a first imaginary signal; The complex convolution layer obtains a first output signal based on a third calculation formula, the first real signal and the first imaginary signal; Among them, the first calculation formula is: The second calculation formula is: The third calculation formula is: In the formula, S i,k represents the first output signal at position i and output channel k, j represents the imaginary unit, represents the first real signal at position i and output channel k, represents the first imaginary signal at position i and output channel k, M represents the length of the neuron, represents the real part signal in the coherent complex valued signal to be processed at position i+m, represents the imaginary signal in the coherent complex-valued signal to be processed at position i+m, represents the real neuron weight at position m and output channel k, represents the imaginary neuron weight at position m and output channel k, represents the real bias at output channel k, Represents the imaginary bias at output channel k.

3. The coherent signal processing method according to claim 2, characterized in that: The complex activation layer is used to perform a nonlinear transformation on the first output signal to obtain a second output signal, including: The complex activation layer performs a nonlinear transformation on the first real signal in the first output signal based on a fourth calculation formula to output a second real signal; The complex activation layer performs a nonlinear transformation on the first imaginary signal in the first output signal based on a fifth calculation formula to output a second imaginary signal; The complex activation layer obtains a second output signal based on the sixth calculation formula, the second real signal and the second imaginary signal; Among them, the fourth calculation formula is: The fifth calculation formula is: The sixth calculation formula is: In the formula, y i,k represents the second output signal at position i and output channel k, represents the second real signal at position i and output channel k, Represents the second imaginary signal at position i and output channel k.

4. The coherent signal processing method according to claim 1, characterized in that: The target class neuron model also includes a complex normalization layer, which is used to normalize the second output signal and input the normalized second output signal to the complex flattening layer.

5. A coherent signal processing device, characterized in that: The coherent signal processing device comprises: An input module, which is used to input the coherent complex-valued signal to be processed into a target-type neuron model, wherein the target-type neuron model includes a complex convolution layer, a complex activation layer, a complex flattening layer, and a complex fully connected layer; A control module, which is used to control the target class neuron model to output a target coherent complex-valued signal based on the coherent complex-valued signal to be processed; wherein the complex convolution layer is used to perform a convolution operation on the coherent complex-valued signal to be processed and the neurons in the complex convolution layer to obtain a first output signal; the complex activation layer is used to perform a nonlinear transformation on the first output signal to obtain a second output signal; the complex flattening layer is used to flatten the second output signal to obtain a target one-dimensional vector; the complex fully connected layer is used to perform feature mapping on the target one-dimensional vector to output a target coherent complex-valued signal.

6. The coherent signal processing device according to claim 5, characterized in that: The complex convolution layer is used to perform a convolution operation on the coherent complex-valued signal to be processed and the neurons in the complex convolution layer based on a first calculation formula to output a first real signal; The complex convolution layer is used to perform a convolution operation on the coherent complex-valued signal to be processed and the neurons in the complex convolution layer based on a second calculation formula to output a first imaginary signal; The complex convolution layer is used to obtain a first output signal based on a third calculation formula, a first real signal and a first imaginary signal; Among them, the first calculation formula is: The second calculation formula is: The third calculation formula is: In the formula, S i,k represents the first output signal at position i and output channel k, j represents the imaginary unit, represents the first real signal at position i and output channel k, represents the first imaginary signal at position i and output channel k, M represents the length of the neuron, represents the real part signal in the coherent complex valued signal to be processed at position i+m, represents the imaginary signal in the coherent complex-valued signal to be processed at position i+m, represents the real neuron weight at position m and output channel k, represents the imaginary neuron weight at position m and output channel k, represents the real bias at output channel k, Represents the imaginary bias at output channel k.

7. The coherent signal processing device according to claim 6, characterized in that: The complex activation layer is used to perform a nonlinear transformation on the first real signal in the first output signal based on a fourth calculation formula to output a second real signal; The complex activation layer is used to perform a nonlinear transformation on the first imaginary signal in the first output signal based on a fifth calculation formula to output a second imaginary signal; The complex activation layer is used to obtain a second output signal based on a sixth calculation formula, a second real signal and a second imaginary signal; Among them, the fourth calculation formula is: The fifth calculation formula is: The sixth calculation formula is: In the formula, y i,k represents the second output signal at position i and output channel k, represents the second real signal at position i and output channel k, Represents the second imaginary signal at position i and output channel k.

8. The coherent signal processing device according to claim 5, characterized in that: The target class neuron model also includes a complex normalization layer, which is used to normalize the second output signal and input the normalized second output signal to the complex flattening layer.

9. A coherent signal processing device, characterized in that: The coherent signal processing device includes a processor, a memory, and a coherent signal processing program stored in the memory and executable by the processor, wherein when the coherent signal processing program is executed by the processor, the steps of the coherent signal processing method as described in any one of claims 1 to 4 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a coherent signal processing program, wherein when the coherent signal processing program is executed by a processor, the steps of the coherent signal processing method according to any one of claims 1 to 4 are implemented.

Images