A fast and high-precision modulation identification method, device, equipment and medium

Abstract

The application provides a fast high-precision modulation recognition method, device, equipment and medium, and the overall scheme design includes two parts of small sample representation technology and multi-modal learning technology. ′ And N ′ is the sample number of the multi-modal learning model input, and the sampling time is reduced. The signal is represented as a sequence and an image form at the same time, and an input adaptive technology is used to solve the input size mismatch problem caused by sample reduction. Through multi-modal learning technology, the feature tensors extracted by the two modes are fused by using a deep neural network, so as to reduce the recognition accuracy loss caused by sample reduction, thereby ensuring the accuracy of recognition.

Description

Technical Field

[0001] This invention relates to the field of communication technology, and specifically to a fast and high-precision modulation identification method, apparatus, device, and medium. Background Technology

[0002] Radio spectrum is a crucial resource in wireless communication systems. To achieve efficient and reliable communication, radio spectrum sensing technology is essential for managing and scheduling spectrum resources. Modulation identification plays a vital role in radio spectrum sensing, particularly in software-defined radio and military applications. By accurately identifying the modulation scheme of received signals, researchers can better understand the characteristics and structure of the signals and adopt appropriate demodulation and processing methods to achieve reliable data transmission and effective communication system performance.

[0003] Deep learning-based modulation recognition (MCR) generally utilizes two types of modal data: sequential modal data and image modal data. Common MCR data for the former includes IQ sequences, amplitude-phase sequences, and Fast Fourier Transform (FFT) sequences. Common image modal data for the latter includes constellation diagrams, eye diagrams, feature point maps, and spectral correlation function images. Numerous studies have demonstrated that deep learning-based MCR achieves good recognition performance across various modal data types. However, research on modulation recognition has not yet fully explored the potential of utilizing multimodal data, despite its superior learning capabilities.

[0004] With the rapid development of communication technology, the communication environment is becoming increasingly complex, and spectrum resources are becoming increasingly limited. Therefore, the demand for modulation recognition is becoming more urgent, requiring faster recognition speeds. However, in deep learning-based modulation recognition systems, signal sampling time is crucial for achieving fast and high-precision modulation recognition. In signal analysis, shortening the sampling time means reducing the number of signal samples, which negatively impacts the accurate identification of modulation schemes; that is, there is a contradiction between shortening the sampling time and improving recognition accuracy. Therefore, the main challenge of existing fast and high-precision modulation recognition methods lies in finding a balance between recognition accuracy and shortening the sampling time, in order to solve the problem of ensuring recognition accuracy while simultaneously achieving rapid sampling.

[0005] In view of the above, this application is hereby submitted. Summary of the Invention

[0006] In view of this, the purpose of the present invention is to provide a fast and high-precision modulation recognition method, apparatus, device and medium, which can effectively solve the problem that the reduction of samples in the existing fast and high-precision modulation recognition methods will lead to a loss of recognition accuracy, thereby affecting the recognition accuracy.

[0007] This invention discloses a fast and high-precision modulation recognition method, comprising:

[0008] The communication signal to be identified is acquired, and the communication signal is subjected to signal sampling processing to generate multiple signal samples;

[0009] Each signal sample is classified to generate an image representation and a sequence representation, wherein the resolution of the image representation is [resolution missing]. The length of the sequence representation is , The width of the image representation. The height represented by the image;

[0010] Resolution of the image representation Adaptive processing is performed to convert and generate multimodal learning image representations from the input resolution. ;

[0011] The length of the sequence representation Adaptive processing is performed to transform and generate a multimodal learning sequence representation of the input length. ;

[0012] The input resolution of the multimodal learning image is represented using a deep neural network. and the multimodal learning sequence characterizes the input length Perform deep feature extraction processing to generate image representation feature vectors and sequence representation feature vectors;

[0013] The image representation feature vector and the sequence representation feature vector are subjected to feature fusion processing to generate a fused feature vector;

[0014] The fused feature vector is fed into a classifier for classification and decision processing to generate a modulation recognition result, wherein the modulation recognition result is the decision result obtained after classifier processing.

[0015] Preferably, the number of the generated signal samples is less than the number of samples used in the signal representation during the training phase.

[0016] Preferably, the resolution of the image representation Adaptive processing is performed to convert and generate multimodal learning image representations from the input resolution. Specifically:

[0017] According to the formula Resolution of the image representation and the resolution of the preset target image Perform calculations to generate a width scaling factor. and height scaling factor ,in, The width of the preset target image, The height of the preset target image;

[0018] According to the formula For the width scaling factor The height scaling factor and the pixel position of the preset target image Perform calculations to generate the pixel positions representing the image. Wherein, the pixel position represented by the image The coordinates of the four nearest neighbor pixels are respectively , , , , The coordinates of the top-left neighboring pixel. The coordinates of the top right neighboring pixel. The coordinates of the bottom left neighboring pixel. The coordinates of the bottom right neighboring pixel;

[0019] According to the formula The pixel positions represented by the image and the coordinates of the upper left corner neighboring pixels Perform calculations to generate the pixel positions representing the image. Offset relative to the top left neighboring pixel ;

[0020] For the coordinates of the upper left neighboring pixels The coordinates of the upper right corner neighboring pixels The coordinates of the lower corner neighboring pixels The coordinates of the lower right corner neighboring pixels and the offset A weighted average calculation is performed to generate the pixel value at the pixel position of the preset target image. ,in, ;

[0021] Repeat the above steps to calculate the pixel values ​​at all preset pixel positions in the preset target image, generating a multimodal learning image representation of the input resolution. The multimodal learning image represents the input resolution. The resolution of the preset target image Consistent.

[0022] Preferably, the length of the sequence representation Adaptive processing is performed to transform and generate a multimodal learning sequence representation of the input length. Specifically:

[0023] According to the formula The length of the sequence representation and the preset target complex sequence length Perform computational processing to generate the length relationship between the sequence representation and the preset target complex sequence. ;

[0024] The sequence representation is spliced ​​multiple times to obtain a complex sequence of length denoted by a preset length value, wherein the preset length value is... ;

[0025] Perform a preprocessing on the complex sequence Length truncation processing to generate multimodal learning sequence representations of the input length The multimodal learning sequence represents the input length. With the preset target complex sequence length Consistent.

[0026] Preferably, when the length relationship If the result is not an integer, round it up.

[0027] The present invention also discloses a fast and high-precision modulation identification device, comprising:

[0028] A signal sampling unit is used to acquire the communication signal to be identified and to perform signal sampling processing on the communication signal to generate multiple signal samples;

[0029] A signal representation unit is used to classify each signal sample, generating image representations and sequence representations, wherein the resolution of the image representation is [resolution missing]. The length of the sequence representation is , The width of the image representation. The height represented by the image;

[0030] Resolution input adaptive unit, used for adjusting the resolution of the image representation. Adaptive processing is performed to convert and generate multimodal learning image representations from the input resolution. ;

[0031] Length input adaptive unit, used for adjusting the length of the sequence representation Adaptive processing is performed to transform and generate a multimodal learning sequence representation of the input length. ;

[0032] The feature extraction unit is used to represent the input resolution of the multimodal learning image using a deep neural network. and the multimodal learning sequence characterizes the input length Perform deep feature extraction processing to generate image representation feature vectors and sequence representation feature vectors;

[0033] The feature fusion unit is used to perform feature fusion processing on the image representation feature vector and the sequence representation feature vector to generate a fused feature vector;

[0034] The classification decision unit is used to send the fused feature vector to the classifier for classification decision processing and generate modulation recognition result, wherein the modulation recognition result is the decision result obtained after classifier processing.

[0035] Preferably, the resolution input adaptive unit is specifically used for:

[0036] According to the formula Resolution of the image representation and the resolution of the preset target image Perform calculations to generate a width scaling factor. and height scaling factor ,in, The width of the preset target image, The height of the preset target image;

[0037] According to the formula For the width scaling factor The height scaling factor and the pixel position of the preset target image Perform calculations to generate the pixel positions representing the image. Wherein, the pixel position represented by the image The coordinates of the four nearest neighbor pixels are respectively , , , , The coordinates of the top-left neighboring pixel. The coordinates of the top right neighboring pixel. The coordinates of the bottom left neighboring pixel. The coordinates of the bottom right neighboring pixel;

[0038] According to the formula The pixel positions represented by the image and the coordinates of the upper left corner neighboring pixels Perform calculations to generate the pixel positions representing the image. Offset relative to the top left neighboring pixel ;

[0039] For the coordinates of the upper left neighboring pixels The coordinates of the upper right corner neighboring pixels The coordinates of the lower corner neighboring pixels The coordinates of the lower right corner neighboring pixels and the offset A weighted average calculation is performed to generate the pixel value at the pixel position of the preset target image. ,in, ;

[0040] Repeat the above steps to calculate the pixel values ​​at all preset pixel positions in the preset target image, generating a multimodal learning image representation of the input resolution. The multimodal learning image represents the input resolution. The resolution of the preset target image Consistent.

[0041] Preferably, the length input adaptive unit is specifically used for:

[0042] According to the formula The length of the sequence representation and the preset target complex sequence length Perform computational processing to generate the length relationship between the sequence representation and the preset target complex sequence. ;

[0043] The sequence representation is spliced ​​multiple times to obtain a complex sequence of length denoted by a preset length value, wherein the preset length value is... ;

[0044] Perform a preprocessing on the complex sequence Length truncation processing to generate multimodal learning sequence representations of the input length The multimodal learning sequence represents the input length. With the preset target complex sequence length Consistent.

[0045] The present invention also discloses a fast and high-precision modulation identification device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements a fast and high-precision modulation identification method as described above.

[0046] The present invention also discloses a readable storage medium storing a computer program, which can be executed by the processor of the device where the storage medium is located to implement a fast and high-precision modulation recognition method as described in any of the above claims.

[0047] In summary, this embodiment provides a fast and high-precision modulation recognition method, apparatus, device, and medium, comprising two parts: small-sample representation technology and multimodal learning technology. The small-sample representation technology is used in the inference stage to perform... Point sampling, where and This reduces the number of input samples to the multimodal learning model and decreases sampling time. The signal is simultaneously represented as both a sequence and an image, and an input adaptation technique is used to address the input size mismatch caused by the reduced sample size. This solves the problem in existing fast and high-precision modulation recognition methods where reduced samples lead to a loss of recognition accuracy, thus affecting the overall recognition accuracy. Attached Figure Description

[0048] Figure 1 This is a flowchart illustrating a fast and high-precision modulation recognition method provided in the first aspect of the present invention.

[0049] Figure 2 This is a flowchart illustrating a fast and high-precision modulation recognition method provided in the second aspect of the present invention.

[0050] Figure 3 This is a schematic diagram illustrating the recognition performance of a fast and high-precision modulation recognition method provided in an embodiment of the present invention.

[0051] Figure 4 This is a schematic diagram of a fast and high-precision modulation recognition device provided in an embodiment of the present invention. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to represent selected embodiments of the invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0054] Please see Figures 1 to 2The first embodiment of the present invention provides a fast and high-precision modulation recognition method, which can be executed by a fast and high-precision modulation recognition device (hereinafter referred to as the recognition device), and in particular, by one or more processors within the recognition device, to achieve the following steps:

[0055] S101, acquire the communication signal to be identified, and perform signal sampling processing on the communication signal to generate multiple signal samples;

[0056] Specifically, in this embodiment, the number of the multiple signal samples generated is less than the number of samples used in the signal representation during the training phase.

[0057] In this embodiment, the fast and high-precision modulation recognition method specifically includes six steps. The first step is signal sampling, which involves sampling the signal to be recognized to obtain... 1 signal sample, of which < and The number of samples used for signal representation during the training phase.

[0058] S102, each signal sample is classified to generate an image representation and a sequence representation, wherein the resolution of the image representation is... The length of the sequence representation is , The width of the image representation. The height represented by the image;

[0059] In this embodiment, the second step is signal representation, which represents the signal in two forms: image and sequence. The output image representation has a resolution of [resolution value missing]. The sequence representation length is .

[0060] S103, the resolution of the image representation Adaptive processing is performed to convert and generate multimodal learning image representations from the input resolution. ;

[0061] Specifically, step S103 includes: according to the formula Resolution of the image representation and the resolution of the preset target image Perform calculations to generate a width scaling factor. and height scaling factor ,in, The width of the preset target image, The height of the preset target image;

[0062] According to the formula For the width scaling factor The height scaling factor and the pixel position of the preset target image Perform calculations to generate the pixel positions representing the image. Wherein, the pixel position represented by the image The coordinates of the four nearest neighbor pixels are respectively , , , , The coordinates of the top-left neighboring pixel. The coordinates of the top right neighboring pixel. The coordinates of the bottom left neighboring pixel. The coordinates of the bottom right neighboring pixel;

[0063] According to the formula The pixel positions represented by the image and the coordinates of the upper left corner neighboring pixels Perform calculations to generate the pixel positions representing the image. Offset relative to the top left neighboring pixel ;

[0064] For the coordinates of the upper left neighboring pixels The coordinates of the upper right corner neighboring pixels The coordinates of the lower corner neighboring pixels The coordinates of the lower right corner neighboring pixels and the offset A weighted average calculation is performed to generate the pixel value at the pixel position of the preset target image. ,in, ;

[0065] Repeat the above steps to calculate the pixel values ​​at all preset pixel positions in the preset target image, generating a multimodal learning image representation of the input resolution. The multimodal learning image represents the input resolution. The resolution of the preset target image Consistent.

[0066] Specifically, in this embodiment, the third step is input adaptation, which is related to the image representation resolution. The input resolution is converted into a multimodal learning image representation through adaptive processing. .

[0067] In this embodiment, it can be divided into four steps. The first step is to assume that the original image size is... ,in Indicates width, Indicates height. Assume the target image size is... Calculate the scaling factor The second step is to assume that the pixel positions of the target image are... This indicates that the pixel positions of the original image are represented by... It means that, among them , Let the coordinates of the four nearest neighbor pixels in the original image be... , , , ,in The coordinates of the top left pixel. The coordinates of the bottom right pixel; calculate Offset relative to the nearest neighbor pixel , The third step is to calculate the target image pixel value by performing a weighted average based on the values ​​of the nearest neighbor pixels and their offsets. ,in, The fourth step is to repeat the third step until the pixel values ​​of all pixel locations in the target image have been processed, thus obtaining the target image. .

[0068] S104, regarding the length of the sequence representation Adaptive processing is performed to transform and generate a multimodal learning sequence representation of the input length. ;

[0069] Specifically, step S104 includes: according to the formula The length of the sequence representation and the preset target complex sequence length Perform computational processing to generate the length relationship between the sequence representation and the preset target complex sequence. ;

[0070] The sequence representation is spliced ​​multiple times to obtain a complex sequence of length denoted by a preset length value, wherein the preset length value is... ;

[0071] Perform a preprocessing on the complex sequence Length truncation processing to generate multimodal learning sequence representations of the input length The multimodal learning sequence represents the input length. With the preset target complex sequence length Consistent.

[0072] In this embodiment, when the length relationship If the result is not an integer, round it up.

[0073] Specifically, in this embodiment, the third step is input adaptation, which is related to the sequence representation length. The input length is represented by a multimodal learning sequence obtained through adaptive processing. .

[0074] In this embodiment, it can be divided into three steps. The first step is to assume that the length of the original complex sequence is... The length of the target complex sequence is Calculate the length relationship between the target complex sequence and the original complex sequence. , Round up. The second step is to process the original complex sequence... After several concatenations, a length of [length] is obtained. The third step is to process a complex sequence of length . Before the complex sequence Length truncation, followed by adaptive length processing to obtain a length of... The target complex sequence.

[0075] S105, using a deep neural network to characterize the input resolution of the multimodal learning image. and the multimodal learning sequence characterizes the input length Perform deep feature extraction processing to generate image representation feature vectors and sequence representation feature vectors;

[0076] Specifically, in this embodiment, the fourth step is feature extraction, which uses a deep neural network to perform deep feature extraction on the image representation and the sequence representation respectively, to obtain the image representation feature vector and the sequence representation feature vector.

[0077] S106, Perform feature fusion processing on the image representation feature vector and the sequence representation feature vector to generate a fused feature vector;

[0078] Specifically, in this embodiment, the fifth step is feature fusion: before classification decision, the extracted image representation feature vector and sequence representation feature vector are fused to obtain a fused feature vector.

[0079] S107, the fused feature vector is fed into a classifier for classification decision processing to generate a modulation recognition result, wherein the modulation recognition result is the decision result obtained after classifier processing.

[0080] Specifically, in this embodiment, the sixth step is to make a classification decision, which involves feeding the fused feature vector into a classifier for classification and decision, and obtaining the decision result as the modulation recognition result.

[0081] In summary, the fast and high-precision modulation recognition method comprises two parts: few-sample representation technology and multimodal learning technology. The few-sample representation technology is used in the inference stage to perform... Point sampling, where and To reduce the number of input samples for the multimodal learning model and decrease sampling time, the signal is simultaneously represented as both a sequence and an image. An input adaptation technique is employed to address the input size mismatch caused by the reduced sample size. Through multimodal learning, a deep neural network is used to fuse the feature tensors extracted from both modalities, mitigating the loss in recognition accuracy due to the reduced sample size and thus ensuring high recognition accuracy.

[0082] The aforementioned fast and high-precision modulation recognition method fully considers the time required for deep learning-based modulation recognition systems to complete the recognition process. By shortening the sampling time to reduce the number of samples and incorporating multimodal feature fusion—two key steps and ideas—it effectively solves the problem of fast and high-precision modulation recognition, achieving crucial progress in addressing signal analysis and processing issues in complex and ever-changing communication systems. Please refer to [link / reference]. Figure 3 As can be seen, the proposed fast and high-precision modulation recognition method achieves fast and high-precision modulation recognition during the inference stage through small-sample representation technology and multimodal learning technology. Simply put, the inference stage employs small-sample representation technology, using a smaller number of samples to represent the signal to be recognized, reducing sampling time and improving recognition speed; and multimodal learning technology, fusing representation results from different modalities, reduces the recognition accuracy loss caused by small samples, ensuring the accuracy of recognition.

[0083] Please see Figure 4 A second embodiment of the present invention provides a fast and high-precision modulation identification device, comprising:

[0084] The signal sampling unit 201 is used to acquire the communication signal to be identified and to perform signal sampling processing on the communication signal to generate multiple signal samples;

[0085] The signal representation unit 202 is used to classify each signal sample and generate image representation and sequence representation, wherein the resolution of the image representation is [resolution missing]. The length of the sequence representation is , The width of the image representation. The height represented by the image;

[0086] Resolution input adaptive unit 203, used for adjusting the resolution of the image representation. Adaptive processing is performed to convert and generate multimodal learning image representations from the input resolution. ;

[0087] Length input adaptive unit 204, used for adjusting the length of the sequence representation Adaptive processing is performed to transform and generate a multimodal learning sequence representation of the input length. ;

[0088] Feature extraction unit 205 is used to represent the input resolution of the multimodal learning image using a deep neural network. and the multimodal learning sequence characterizes the input length Perform deep feature extraction processing to generate image representation feature vectors and sequence representation feature vectors;

[0089] The feature fusion unit 206 is used to perform feature fusion processing on the image representation feature vector and the sequence representation feature vector to generate a fused feature vector.

[0090] The classification decision unit 207 is used to send the fused feature vector to the classifier for classification decision processing to generate a modulation recognition result, wherein the modulation recognition result is the decision result obtained after the classifier processing.

[0091] Preferably, the resolution input adaptive unit 203 is specifically used for:

[0092] According to the formula Resolution of the image representation and the resolution of the preset target image Perform calculations to generate a width scaling factor. and height scaling factor ,in, The width of the preset target image, The height of the preset target image;

[0093] According to the formula For the width scaling factor The height scaling factor and the pixel position of the preset target image Perform calculations to generate the pixel positions representing the image. Wherein, the pixel position represented by the image The coordinates of the four nearest neighbor pixels are respectively , , , , The coordinates of the top-left neighboring pixel. The coordinates of the top right neighboring pixel. The coordinates of the bottom left neighboring pixel. The coordinates of the bottom right neighboring pixel;

[0094] According to the formula The pixel positions represented by the image and the coordinates of the upper left corner neighboring pixels Perform calculations to generate the pixel positions representing the image. Offset relative to the top left neighboring pixel ;

[0095] For the coordinates of the upper left neighboring pixels The coordinates of the upper right corner neighboring pixels The coordinates of the lower corner neighboring pixels The coordinates of the lower right corner neighboring pixels and the offset A weighted average calculation is performed to generate the pixel value at the pixel position of the preset target image. ,in, ;

[0096] Repeat the above steps to calculate the pixel values ​​at all preset pixel positions in the preset target image, generating a multimodal learning image representation of the input resolution. The multimodal learning image represents the input resolution. The resolution of the preset target image Consistent.

[0097] Preferably, the length input adaptive unit 204 is specifically used for:

[0098] According to the formula The length of the sequence representation and the preset target complex sequence length Perform computational processing to generate the length relationship between the sequence representation and the preset target complex sequence. ;

[0099] The sequence representation is spliced ​​multiple times to obtain a complex sequence of length denoted by a preset length value, wherein the preset length value is... ;

[0100] Perform a preprocessing on the complex sequence Length truncation processing to generate multimodal learning sequence representations of the input length The multimodal learning sequence represents the input length. With the preset target complex sequence length Consistent.

[0101] A third embodiment of the present invention provides a fast and high-precision modulation identification device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements a fast and high-precision modulation identification method as described above.

[0102] A fourth embodiment of the present invention provides a readable storage medium storing a computer program, which can be executed by the processor of the device where the storage medium is located to implement a fast and high-precision modulation recognition method as described in any of the above embodiments.

[0103] Exemplary examples show that the computer program described in the third and fourth embodiments of the present invention can be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules can be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in implementing a fast and high-precision modulation identification device. For example, the apparatus described in the second embodiment of the present invention.

[0104] The processor referred to can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the fast, high-precision modulation identification method, connecting various parts of the method through various interfaces and lines.

[0105] The memory can be used to store the computer program and / or modules. The processor, by running or executing the computer program and / or modules stored in the memory, and by calling the data stored in the memory, implements various functions of a fast and high-precision modulation recognition method. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, text conversion function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, text message data, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0106] If the implemented module is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0107] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0108] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions that fall within the scope of the present invention are within the scope of protection of the present invention.

Claims

1. A fast high-precision modulation identification method, characterized in that, The method comprises the following steps: acquiring a communication signal to be identified, and performing signal sampling processing on the communication signal to generate a plurality of signal samples; Classifying each of the signal samples to generate an image representation and a sequence representation, wherein a resolution of the image representation is , a length of the sequence representation is , a width of the image representation, a height of the image representation; resolution of the image representation performing adaptive processing, converting to generate a multi-modal learning image representation input resolution ; Length of the sequence representation Adaptive processing is performed to convert the generated multi-modal learning sequence representation input length ; performing deep feature extraction processing on the multi-modal learning image representation input resolution and the multi-modal learning sequence representation input length to generate image representation feature vectors and sequence representation feature vectors; performing feature fusion processing on the image representation feature vector and the sequence representation feature vector to generate a fusion feature vector; performing classification decision processing on the fusion feature vector in a classifier to generate a modulation recognition result, wherein the modulation recognition result is a decision result obtained after the processing of the classifier; a resolution of the image representation performing adaptive processing to convert to generate a multi-modal learning image representation input resolution , specifically: According to the formula a resolution of the image and a resolution of the preset target image a calculation process is performed to generate a width scaling factor and a height scaling factor wherein, the width of the preset target image, the height of the preset target image; According to the formula The width scaling factor , the height scaling factor and the pixel position of the preset target image are calculated and processed to generate the pixel position of the image representation , wherein the four nearest neighbor pixel coordinates of the pixel position of the image representation are respectively , , , , is the upper left corner neighbor pixel coordinate, is the upper right corner neighbor pixel coordinate, is the lower left corner neighbor pixel coordinate, is the lower right corner neighbor pixel coordinate. According to the formula the pixel position of the image representation and the upper left corner neighboring pixel coordinates a calculation process is performed to generate the pixel position of the image representation relative to the offset of the upper left corner neighboring pixel ; performing a weighted average calculation process on the left upper corner neighboring pixel coordinates , the right upper corner neighboring pixel coordinates , the lower corner neighboring pixel coordinates , the right lower corner neighboring pixel coordinates , and the offset to generate a pixel value of a pixel position of the preset target image , wherein ; Repeat the above steps to calculate the pixel values of all preset pixel positions in the preset target image, generate a multi-modal learning image representation input resolution , wherein the multi-modal learning image representation input resolution is consistent with the resolution of the preset target image.

2. The method of claim 1, wherein, the number of the generated plurality of signal samples is smaller than the number of samples used in the signal representation in the training phase.

3. The method of claim 1, wherein the method is a fast and high-precision modulation identification method. Length of the sequence representation Adaptive processing is performed to convert the generated multi-modal learning sequence representation input length Specifically: According to the formula The length of the sequence is characterized And the preset target complex sequence length The length relationship between the sequence and the preset target complex sequence is generated by calculation processing ; The sequence representation is spliced multiple times to obtain a plurality of sequences with a preset length value, wherein the preset length value is ; Performing a length truncation processing on the complex sequence to generate a multimodal learning sequence representation input length The length truncation processing is performed on the complex sequence to generate a multimodal learning sequence representation input length The multimodal learning sequence representation input length is consistent with the preset target complex sequence length The multimodal learning sequence representation input length is consistent with the preset target complex sequence length The multimodal learning sequence representation input length is consistent with the preset target complex sequence length 4. The method of claim 3, wherein the method is a fast and high-precision modulation identification method. When the length relationship Rounded up if not an integer.

5. A fast high-precision modulation identification device, characterized in that, The method comprises the following steps: a signal sampling unit configured to acquire a communication signal to be identified, and perform signal sampling processing on the communication signal to generate a plurality of signal samples; The signal representation unit is configured to classify each of the signal samples to generate an image representation and a sequence representation, wherein a resolution of the image representation is , and a length of the sequence representation is , is a width of the image representation, is a height of the image representation. a resolution input adaptive unit for adapting the resolution of the image representation performing an adaptive processing, converting the generated multi-modal learning image representation input resolution ; a length input adaptive unit configured to adapt a length of the sequence representation adaptively process the multi-modal learning sequence representation input length ; The feature extraction unit is used to represent the input resolution of the multimodal learning image using a deep neural network. and the multimodal learning sequence characterizes the input length Perform deep feature extraction processing to generate image representation feature vectors and sequence representation feature vectors; a feature fusion unit configured to perform feature fusion processing on the image representation feature vector and the sequence representation feature vector to generate a fusion feature vector; a classification decision unit configured to perform classification decision processing on the fusion feature vector in a classifier to generate a modulation recognition result, wherein the modulation recognition result is a decision result obtained after the processing of the classifier; The resolution input adaptive unit is specifically configured to: According to the formula the resolution of the image and the resolution of the preset target image a calculation process is performed to generate a width scaling factor and a height scaling factor wherein, the width of the preset target image, the height of the preset target image; According to the formula The width scaling factor , the height scaling factor and the pixel position of the preset target image are calculated and processed to generate the pixel position of the image representation , wherein the four nearest neighbor pixel coordinates of the pixel position of the image representation are respectively , , , , is the upper left corner neighbor pixel coordinate, is the upper right corner neighbor pixel coordinate, is the lower left corner neighbor pixel coordinate, is the lower right corner neighbor pixel coordinate. According to the formula the pixel position of the image representation and the upper left corner neighboring pixel coordinates a calculation process is performed to generate the pixel position of the image representation relative to the offset of the upper left corner neighboring pixel ; performing a weighted average calculation process on the left upper corner neighboring pixel coordinates , the right upper corner neighboring pixel coordinates , the lower corner neighboring pixel coordinates , the right lower corner neighboring pixel coordinates , and the offset to generate a pixel value of a pixel position of the preset target image , wherein ; Repeat the above steps to calculate the pixel values of all preset pixel positions in the preset target image, generate a multi-modal learning image representation input resolution , wherein the multi-modal learning image representation input resolution is consistent with the resolution of the preset target image.

6. A fast high precision modulation recognition apparatus as claimed in claim 5, characterized in that The length input adaptive unit is specifically configured to: According to the formula The length of the sequence representation And the preset target complex sequence length Carrying out calculation processing, generating the length relationship between the sequence representation and the preset target complex sequence ; The sequence representation is spliced multiple times to obtain a plurality of sequences with a preset length value, wherein the preset length value is ; Performing a length truncation processing on the complex sequence to generate a multimodal learning sequence representation input length The length truncation processing is performed on the complex sequence to generate a multimodal learning sequence representation input length The multimodal learning sequence representation input length is consistent with the preset target complex sequence length The multimodal learning sequence representation input length is consistent with the preset target complex sequence length The multimodal learning sequence representation input length is consistent with the preset target complex sequence length 7. A fast high-precision modulation identification device, characterized by, The computer program is configured to be executed by the processor, and the processor implements the fast and high-precision modulation recognition method according to any one of claims 1 to 4 when executing the computer program.

8. A readable storage medium, characterized by, The computer program can be executed by the processor of the device where the storage medium is located to implement the fast and high-precision modulation recognition method according to any one of claims 1 to 4.

Images