A lightweight channel-coded blind identification method and system based on multimodal feature fusion

Abstract

The application belongs to the technical field of wireless communication, and specifically discloses a light-weight channel coding blind identification method and system based on multi-modal feature fusion. First, the signal-to-noise ratio of the collected signal is detected. If the signal-to-noise ratio is less than a signal-to-noise ratio threshold, the collected signal is subjected to LDPC code identification, Polar code identification, Turbo code identification, convolution code identification and linear block code identification based on the theory of mathematical statistics. Otherwise, the collected signal is input into a channel coding identification model for blind identification of channel coding. The method of the application adaptively selects signal statistical features and signal depth features for blind identification based on the signal-to-noise ratio of the collected signal, and combines the high efficiency of mathematical statistics identification and the high precision of deep learning identification, thereby achieving a dynamic balance between the calculation efficiency and identification precision of signal blind identification.

Description

Technical Field

[0001] This application belongs to the field of wireless communication technology, and more specifically, relates to a lightweight channel coding blind identification method and system based on multimodal feature fusion. Background Technology

[0002] Blind channel coding identification refers to the process of inferring the channel coding type and parameters (such as code length, code rate, generator polynomial, etc.) from the received signal without knowing the coding parameters. It is of great significance in non-cooperative communication and signal analysis.

[0003] Currently, blind channel coding identification techniques are mainly divided into two categories. One category is channel coding identification methods based on mathematical statistics theory. This method analyzes the periodic components of the received signal and manually extracts coding features to identify the channel code. Traditional channel identification algorithms can usually only identify specific channel codes, making it difficult to identify channel code types in complex electromagnetic environments, and the algorithms are also highly complex. The other category is deep learning methods based on data accumulation. This method labels known data, typically using the original bitstream as input, and extracts high-dimensional features from the original bitstream through a convolutional neural network to distinguish different channel coding types. However, its problem is that the model has a large number of parameters, making it difficult to deploy on edge devices for real-time identification. Summary of the Invention

[0004] In view of the above-mentioned defects or improvement needs of the existing technology, this application provides a lightweight channel coding blind identification method and system based on multimodal feature fusion, which aims to solve the technical problem that the existing channel coding blind identification is too complicated.

[0005] To achieve the above objectives, in a first aspect, this application provides a lightweight channel coding blind identification method based on multimodal feature fusion, comprising:

[0006] The signal-to-noise ratio (SNR) of the acquired signal is detected. If the SNR is less than a threshold, LDPC code recognition, Polar code recognition, Turbo code recognition, convolutional code recognition, and linear block code recognition are performed on the acquired signal based on signal statistical characteristics. Otherwise, the acquired signal is input into a channel coding recognition model for blind channel coding recognition. The channel coding recognition model is trained using the following steps:

[0007] A dataset including LDPC code signals, Polar code signals, Turbo code signals, convolutional code signals, and linear block code signals was generated through simulation.

[0008] High-frequency noise and low-frequency noise are added proportionally to each type of simulated signal in the dataset, and the amplitude scaling factor of each type of simulated signal is adaptively adjusted according to the signal power.

[0009] The dataset for each type of simulated signal is divided into a training set and a test set according to the proportions.

[0010] A channel coding recognition model based on a convolutional neural network is constructed. All training sets are input into the channel coding recognition model for training. Finally, all test sets are input into the trained and converged model for testing, resulting in a well-trained channel coding recognition model.

[0011] The preferred option is:

[0012] Pre-construct candidate parity-check matrices for LDPC codes;

[0013] If the average posterior log-likelihood ratio of the parity check matrix and the acquired signal is higher than a preset first threshold, the acquired signal is determined to be an LDPC code, and the LDPC code parameters are identified based on the candidate parity check matrix.

[0014] Preferably, the average posterior log-likelihood ratio of the verification matrix and the acquired signal is as follows:

[0015]

[0016] in, For the candidate verification matrix, In order to collect signals, To collect signals The Middle bits, To collect signals The total number of bits in the data; for and The posterior log-likelihood ratio.

[0017] Preferably, the acquired signal is subjected to Polar code recognition, specifically as follows:

[0018] A sliding window is used to capture the acquired signal and obtain multiple sets of codewords;

[0019] Constructing a matrix from multiple sets of codewords and judge If the condition is met, the acquired signal is determined to be a Polar code, and the Polar code parameters are identified; where... Representation matrix It is OK A complex matrix of columns, For the number of code words, For candidate code length, Represents finding a matrix Rank.

[0020] Preferably, the acquired signal is subjected to Turbo code recognition, specifically as follows:

[0021] Constructing the Hankel matrix based on the acquired signals ;

[0022] Traversing candidate values ​​of different code lengths If it can be satisfied If so, the acquired signal is determined to be a Turbo code, and This is the code length of the Turbo code.

[0023] Preferably, convolutional code recognition is performed on the acquired signal, specifically as follows:

[0024] The acquired signal is detected using a cyclic spectrum correlation function. If the cyclic frequency can be detected, the acquired signal is determined to be a convolutional code, and the convolutional code parameters are identified based on the cyclic frequency.

[0025] Preferably, a cyclic spectrum correlation function is used to detect the acquired signal, specifically as follows:

[0026]

[0027] in, In order to collect signals, For time delay, It is the duration of the analysis window. The base of the natural logarithm, It is pi. It is the cycle frequency. It is the symbol for imaginary numbers.

[0028] Preferably, the acquired signal is subjected to linear block code recognition, specifically as follows:

[0029] Preconstruct candidate codewords for linear block codes;

[0030] The code likelihood of the acquired signal is calculated using the following formula. :

[0031]

[0032] in, For the first One candidate codeword, For the first signal in the acquisition bits, The code length of the acquired signal; The log-likelihood ratio;

[0033] If code likelihood exists If the value is greater than a preset second threshold, the acquired signal is determined to be a linear block code, and the code likelihood is determined accordingly. Identify linear block code parameters.

[0034] Preferably, the amplitude scaling factor for each type of simulated signal is adaptively adjusted according to the signal power, specifically as follows:

[0035]

[0036] It is the signal after scaling. It is the original signal. yes The mean, It is the noise variance. yes variance It is the square root of the signal-to-noise ratio.

[0037] Secondly, this application provides a lightweight channel-coded blind identification system based on multimodal feature fusion, comprising:

[0038] The signal-to-noise ratio (SNR) detection module is used to detect the SNR of the acquired signal. If the SNR is less than the SNR threshold, the signal statistical feature recognition module is called; otherwise, the model recognition module is called.

[0039] The signal statistical feature recognition module is used to perform LDPC code recognition, Polar code recognition, Turbo code recognition, convolutional code recognition, and linear block code recognition on the acquired signal based on the signal statistical features.

[0040] The model recognition module is used to input the acquired signal into the channel coding recognition model for blind channel coding recognition;

[0041] The model training module is used to generate datasets including LDPC code signals, Polar code signals, Turbo code signals, convolutional code signals, and linear block code signals through simulation; to add high-frequency noise and low-frequency noise to each type of simulated signal in the dataset proportionally, and to adaptively adjust the amplitude scaling factor of each type of simulated signal according to the signal power; to divide the dataset of each type of simulated signal into training set and test set proportionally; to construct a channel coding recognition model based on convolutional neural network, to input all training sets into the channel coding recognition model for model training, and finally to input all test sets into the trained and converged model for testing, thereby obtaining the trained channel coding recognition model.

[0042] Overall, the technical solutions conceived in this application have the following beneficial effects compared with the prior art:

[0043] (1) The channel coding blind identification method of this application adaptively selects the statistical features and signal depth features of the signal based on the signal-to-noise ratio of the collected signal for blind identification. Under low signal-to-noise ratio conditions, mathematical statistical identification algorithm is used to reduce computational complexity, and under high signal-to-noise ratio conditions, deep learning identification method is used to improve identification accuracy. Thus, a dynamic balance between computational efficiency and identification accuracy of blind signal identification is achieved.

[0044] (2) The channel coding recognition model constructed by deep learning in this application employs a noise training strategy during training to ensure that the recognition accuracy is greater than 90% under high signal-to-noise ratio conditions.

[0045] (3) The coded blind identification method based on signal statistical features proposed in this application is more suitable for low signal-to-noise ratio conditions and has lower computational complexity than traditional coded blind identification methods. Attached Figure Description

[0046] Figure 1 This application provides a flowchart of a lightweight channel coding blind identification method based on multimodal feature fusion.

[0047] Figure 2 This is a schematic diagram illustrating the trend of loss rate changes during the training process of the channel coding identification model provided in this application embodiment.

[0048] Figure 3 This is a schematic diagram illustrating the changing trend of recognition accuracy during the training process of the channel coding recognition model provided in this application embodiment.

[0049] Figure 4 This is a schematic diagram of the composition structure of a lightweight channel coding blind identification system based on multimodal feature fusion provided in this application embodiment. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0051] The terms "first" and "second," etc., used in the specification and claims herein are used to distinguish different objects, not to describe a specific order of objects. For example, "first codeword" and "second codeword," etc., are used to distinguish different codewords, not to describe a specific order of codewords.

[0052] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0053] In the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more, for example, multiple codewords means two or more codewords, etc.

[0054] First, the technical terms involved in the embodiments of this application will be introduced.

[0055] LDPC (Low-Density Parity-Check Code) is a high-performance linear block error-correcting code with error-correcting capabilities close to the Shannon limit and efficient decoding process.

[0056] Polar codes are a type of linear block code based on the phenomenon of channel polarization. Through the channel polarization process, Polar codes can synthesize multiple parallel independent channels into perfect channels with a capacity close to 1 and pure noise channels with a capacity close to 0, and then directly transmit information on the perfect channels to approximate the channel capacity.

[0057] Turbo codes (concatenated codes) are a type of parallel concatenated convolutional code. The core idea is to improve the randomness and error correction capability of the encoding by connecting multiple convolutional codes in parallel and inserting interleavers between them.

[0058] A Hankel matrix is ​​a special type of matrix in which the elements on each diagonal are identical.

[0059] The embodiments of this application are described below with reference to the accompanying drawings.

[0060] This application discloses a lightweight channel coding blind identification method based on multimodal feature fusion, such as... Figure 1 As shown, the specific steps include:

[0061] Step 1: Acquire signals.

[0062] Step 2: Detect the signal-to-noise ratio of the acquired signal.

[0063]

[0064] in, The signal-to-noise ratio of the acquired signal; When there is a signal input, measure the voltage value at the receiver output terminal; To measure the noise voltage value at the receiver output when there is no signal input.

[0065] Step 3: Adaptively select signal statistical features or signal depth features based on the signal-to-noise ratio for blind signal identification:

[0066] Step 31: Compare the signal-to-noise ratio (SNR) and the SNR threshold. If the SNR is less than the SNR threshold, proceed to step 32; otherwise, proceed to step 33.

[0067] Step 32: Based on mathematical statistics theory, perform LDPC code recognition, Polar code recognition, Turbo code recognition, convolutional code recognition, and linear block code recognition on the acquired signals respectively:

[0068] Step 321: Perform LDPC code recognition on the acquired signal:

[0069] Pre-construct candidate parity-check matrix for LDPC codes ;

[0070] Derive the verification matrix and acquisition signal Mean posterior log-likelihood ratio:

[0071]

[0072] in, For the candidate verification matrix, In order to collect signals, To collect signals The Middle bits, To collect signals The total number of bits in the data; for and The posterior log-likelihood ratio.

[0073] If the average posterior log-likelihood ratio of the parity check matrix and the acquired signal is higher than a preset first threshold, the acquired signal is determined to be an LDPC code, and the LDPC code parameters are identified based on the candidate parity check matrix.

[0074] Taking a Gaussian channel as an example:

[0075]

[0076] It is a natural exponential function. The power of channel noise. For the first One acquired signal, The actual value of the code element; It indicates a direct proportion.

[0077] Step 322: Perform Polar code recognition on the acquired signal:

[0078] Polar code length Typically, it is a power of 2. A sliding window truncation analysis is performed on the acquired signal. If it is a Polar code, the code length corresponding to the peak autocorrelation of the acquired signal satisfies... .

[0079] A sliding window is used to capture the acquired signal and obtain multiple sets of codewords;

[0080] Constructing a matrix from multiple sets of codewords and judge If the condition is met, the acquired signal is determined to be a Polar code, and the Polar code parameters are identified; where... Representation matrix It is OK A complex matrix of columns, For the number of code words, For candidate code length, Represents finding a matrix Rank.

[0081] Step 323: Perform Turbo code recognition on the acquired signal:

[0082] Constructing the Hankel matrix based on the acquired signals ;

[0083]

[0084] A Hankel matrix is ​​a special type of matrix, defined by the requirement that all elements on the antidiagonal are equal, i.e., satisfying the following relation:

[0085]

[0086] Assume the sequence of acquired signals is Determine the number of rows in the Hankel matrix. Number of columns and meet the conditions And fill it in according to the following rules:

[0087]

[0088]

[0089]

[0090] Traversing candidate values ​​of different code lengths If it can be satisfied If so, the acquired signal is determined to be a Turbo code, and This is the code length of the Turbo code.

[0091] Step 324: Perform convolutional code recognition on the acquired signal:

[0092] The acquired signal is detected using a cyclic spectrum correlation function. If the cyclic frequency can be detected, the acquired signal is determined to be a convolutional code, and the convolutional code parameters are identified based on the cyclic frequency.

[0093] The cyclic spectrum correlation function is as follows:

[0094]

[0095] in, In order to collect signals, For time delay, It is the duration of the analysis window. The base of the natural logarithm, It is pi. It is the cycle frequency. It is the symbol for imaginary numbers.

[0096] Step 325: Perform linear block code identification on the acquired signal:

[0097] Preconstruct candidate codewords for linear block codes;

[0098] The code likelihood of the acquired signal is calculated using the following formula. :

[0099]

[0100] in, For the first One candidate codeword, For the first signal in the acquisition bits, The code length of the acquired signal; The log-likelihood ratio;

[0101] If code likelihood exists If the value is greater than a preset second threshold, the acquired signal is determined to be a linear block code, and the code likelihood is determined accordingly. Identify linear block code parameters.

[0102] Step 33: Input the acquired signal into the channel coding identification model for blind identification of channel coding.

[0103] In this embodiment of the application, the channel coding identification model is trained using the following steps:

[0104] Step 1: Generate a dataset including LDPC code signals, Polar code signals, Turbo code signals, convolutional code signals, and linear block code signals through simulation;

[0105] Step 11: Generate LDPC code signals through simulation, specifically as follows:

[0106] From the equation Derive the generator matrix of LDPC code ;in, This is the matrix transpose. Indicates modulo; The parity-check matrix for the LDPC code is generated by randomly constructing a sparse matrix. The number of non-zero elements in each row and each column is fixed, satisfying the low-density characteristic;

[0107] Then from the equation Generate LDPC code; where, The generated LDPC code has a length of ,Include Information bits and One check bit All are positive integers; The original information bit sequence;

[0108] Step 12: Generate Polar code signals through simulation, specifically as follows:

[0109] The formula for generating Polar codes is as follows:

[0110]

[0111] in, This is the encoded Polar code sequence. Given the input information bit sequence, For sequence length, The generator matrix of the Polar code is constructed using the following formula:

[0112]

[0113] in, The basic kernel matrix; Indicates to conduct The recursive operation of the Kronecker product, by equation Push ; It is a bit-reversed permutation matrix used to adjust the order of the input sequence;

[0114] Step 13: Generate Turbo code signals through simulation, specifically as follows:

[0115] Turbo code signal is ;in This is the original information sequence; This is the first verification sequence; This is the second verification sequence;

[0116] For the input original information sequence The RSC encoder will output the original information as And check bits , No. Validation sequence The calculation process is shown in the following formula:

[0117]

[0118] in, For constraint length; Represents the XOR operation; For the binary representation of the feedback polynomial, It is the binary representation of the feedforward polynomial; Indicates modulo; For the first Original information;

[0119] Step 14: Generate convolutional code signals through simulation, specifically as follows:

[0120] For bitrate The constraint length is The convolutional code, its encoding process is composed of There are 1 generator polynomial definition, and each generator polynomial corresponds to one output bit:

[0121]

[0122] in, For the first The first moment One output bit; for Input bits at time t; To generate polynomial coefficients, representing the first polynomial coefficient... Is the output consistent with...? The values ​​of the stage shift registers are connected; Indicates modulo; and It is a positive integer;

[0123] Step 15: Generate a linear block code signal through simulation, specifically as follows:

[0124] Choose the generator matrix of the linear block code Input information vector With matrix Perform matrix multiplication to obtain channel-coded data. ; through the verification matrix ,verify ;in, and It is a positive integer. It is the transpose of the matrix.

[0125] Step 2: Add high-frequency noise and low-frequency noise to each type of simulated signal in the dataset proportionally, and adaptively adjust the amplitude scaling factor of each type of simulated signal according to the signal power. The scaling formula is as follows;

[0126]

[0127] It is the signal after scaling. It is the original signal. yes The mean, It is the noise variance. yes variance It is the square root of the signal-to-noise ratio.

[0128] Step 3: Extract 1024 consecutive data points for each type of simulated signal and divide them into training and test sets in an 8:2 ratio;

[0129] Step 4: Construct a channel coding recognition model based on a convolutional neural network, and set the model training parameters as follows:

[0130] Training batches: 200;

[0131] Batch size: 64;

[0132] Learning rate: 0.01;

[0133] Loss function: Cross-entropy loss function;

[0134] Optimizer: Stochastic gradient descent.

[0135] All training sets are input into the channel coding recognition model for model training. During the training process, such as... Figure 2 As shown, the loss value converges after 200 iterations, completing model training. Figure 3 As shown, during the training process, the model's recognition accuracy approaches its optimal stability after 200 iterations.

[0136] Finally, all test sets are input into the trained and converged model for testing, and the trained channel coding recognition model is obtained.

[0137] This application also discloses a lightweight channel coding blind identification system based on multimodal feature fusion, such as... Figure 4 As shown, it specifically includes:

[0138] The signal-to-noise ratio (SNR) detection module is used to detect the SNR of the acquired signal. If the SNR is less than the SNR threshold, the signal statistical feature recognition module is called; otherwise, the model recognition module is called.

[0139] The signal statistical feature recognition module is used to perform LDPC code recognition, Polar code recognition, Turbo code recognition, convolutional code recognition, and linear block code recognition on the acquired signal based on the signal statistical features.

[0140] The model recognition module is used to input the acquired signal into the channel coding recognition model for blind channel coding recognition;

[0141] The model training module is used to generate a dataset including LDPC code signals, Polar code signals, Turbo code signals, convolutional code signals, and linear block code signals through simulation; add noise to each type of simulated signal in the dataset, and adaptively adjust the amplitude scaling factor of each type of simulated signal according to the signal power; construct a channel coding recognition model based on a convolutional neural network, and train the channel coding recognition model using the dataset.

[0142] It should be understood that the above system is used to execute the methods in the above embodiments. The corresponding program modules in the system are similar in implementation principle and technical effect to those described in the above methods. The working process of the system can be referred to the corresponding process in the above methods, and will not be repeated here.

[0143] Based on the methods in the above embodiments, this application provides an electronic device including a processor, a communications interface, a memory, and a communication bus, wherein the processor, communications interface, and memory communicate with each other via the communication bus. The processor can call logical instructions in the memory to execute the methods in the above embodiments.

[0144] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.

[0145] Based on the methods in the above embodiments, this application provides a computer-readable storage medium storing a computer program that, when run on a processor, causes the processor to execute the methods in the above embodiments.

[0146] Based on the methods in the above embodiments, this application provides a computer program product that, when run on a processor, causes the processor to execute the methods in the above embodiments.

[0147] It is understood that the processor in the embodiments of this application 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, transistor logic devices, hardware components, or any combination thereof. A general-purpose processor can be a microprocessor or any conventional processor.

[0148] The method steps in this application embodiment can be implemented in hardware or by a processor executing software instructions. The software instructions can consist of corresponding software modules, which can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, portable hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can reside in an ASIC.

[0149] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted through the computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state disk (SSD)).

[0150] It is understood that the various numerical designations used in the embodiments of this application are merely for the convenience of description and are not intended to limit the scope of the embodiments of this application.

[0151] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A light-weight channel coding blind recognition method based on multi-modal feature fusion, characterized in that, The method comprises the following steps: detecting the signal-to-noise ratio of the collected signal, if the signal-to-noise ratio is less than a signal-to-noise ratio threshold, performing LDPC code identification, Polar code identification, Turbo code identification, convolutional code identification and linear block code identification on the collected signal based on signal statistical characteristics, otherwise, inputting the collected signal into a channel coding identification model to perform blind identification of channel coding; The channel coding identification model is obtained by the following steps: generating a data set including LDPC code signals, Polar code signals, Turbo code signals, convolutional code signals and linear block code signals through simulation; adding high-frequency noise and low-frequency noise to each type of simulation signal in the data set in proportion, and adaptively adjusting the amplitude scaling factor of each type of simulation signal according to the signal power; dividing the data set of each type of simulation signal into a training set and a test set in proportion; constructing a channel coding identification model based on a convolutional neural network, inputting all the training set into the channel coding identification model for model training, and finally inputting all the test set into the trained model for testing to obtain the trained channel coding identification model.

2. The method of claim 1, wherein The collected signal is subjected to LDPC code identification, specifically: a candidate check matrix of the LDPC code is constructed in advance; if the average posterior log-likelihood ratio of the check matrix and the collected signal is higher than a preset first threshold, it is determined that the collected signal is an LDPC code, and the LDPC code parameters are identified based on the candidate check matrix.

3. The method of claim 2, wherein The average posterior log-likelihood ratio of the check matrix and the collected signal is specifically: wherein, is a candidate check matrix, is a collected signal, is a collected signal in the bit, is a collected signal in the total number of bits; is and the a posteriori log-likelihood ratio.

4. The method of claim 1, wherein The collected signal is subjected to Polar code identification, specifically: a sliding window is used to intercept the collected signal to obtain multiple groups of code words; Constructing a matrix from multiple groups of code words and determining whether the following is true: is true, then it is determined that the collected signal is a Polar code, and Polar code parameters are identified; wherein represents a matrix is a row column complex matrix, is the number of code words, is a candidate code length, represents the rank of the matrix .

5. The method of claim 1, wherein The collected signal is subjected to Turbo code identification, specifically: Constructing a Hankel matrix based on the collected signals ; Traverse different code length candidate values , if the condition is satisfied, the collected signal is determined as a Turbo code, and is the code length of the Turbo code.

6. The method of claim 1, wherein The collected signal is subjected to convolutional code identification, specifically: The collected signal is subjected to convolutional code identification, specifically:

7. The method of claim 6, wherein The collected signal is subjected to convolutional code identification, specifically: wherein is the acquisition signal, is the time delay, is the time length of the analysis window, is the base of the natural logarithm, is the circle constant, is the cycle frequency, is the imaginary unit.

8. The method of claim 1, wherein The collected signal is subjected to linear block code identification, specifically: a candidate check matrix of the LDPC code is constructed in advance; The code pattern likelihood of the acquired signal is calculated using the following equation : wherein, is the th candidate codeword, is the th bit in the collected signal, is the code length of the collected signal; is the log-likelihood ratio; If code likelihood exists If the value is greater than a preset second threshold, the acquired signal is determined to be a linear block code, and the code likelihood is determined accordingly. Identify linear block code parameters.

9. The method of claim 1, wherein and adaptively adjusting the amplitude scaling factor of each type of simulation signal according to the signal power, specifically: is the scaled signal, is the original signal, is the mean of is the noise variance, is the variance of is the square root of the signal-to-noise ratio.

10. A light-weight channel coding blind recognition system based on multi-modal feature fusion, characterized in that, The method comprises the following steps: The signal-to-noise ratio detection module is used for detecting the signal-to-noise ratio of the collected signal, if the signal-to-noise ratio is less than a signal-to-noise ratio threshold, the signal statistical feature identification module is called, otherwise the model identification module is called; The signal statistical feature identification module is used for performing LDPC code identification, Polar code identification, Turbo code identification, convolutional code identification and linear block code identification on the collected signal based on signal statistical characteristics; The model identification module is used for inputting the collected signal into a channel coding identification model to perform blind identification of channel coding; The model training module is configured to generate a data set including LDPC code signals, Polar code signals, Turbo code signals, convolutional code signals and linear block code signals through simulation; add high-frequency noise and low-frequency noise to each type of simulation signal in the data set in proportion, and adaptively adjust the amplitude scaling coefficient of each type of simulation signal according to the signal power; divide the data set of each type of simulation signal into a training set and a test set in proportion; construct a channel coding recognition model based on a convolutional neural network, input all the training set to the channel coding recognition model for model training, finally input all the test set to the model after training convergence for testing, and obtain the trained channel coding recognition model.

Images