Protocol identification and parameter analysis method and device for wireless communication signal based on deep learning
By automatically identifying communication protocols and parsing signal parameters through a cascaded ViT model, the inefficiency of traditional methods is solved, enabling efficient and accurate communication signal analysis and supporting a comprehensive wireless communication signal tester for multiple application scenarios.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- ANHUI UNIV
- Filing Date
- 2026-03-26
- Publication Date
- 2026-05-15
AI Technical Summary
Traditional communication protocol analysis methods require expensive equipment and expertise, are inefficient, cannot achieve real-time response, and existing deep learning models struggle to balance accuracy and computational efficiency in protocol identification and parameter parsing.
A multimodal signal recognition model is adopted, and the communication protocol and parameters are automatically identified and parsed through cascaded ViT models. This includes data preprocessing, image segmentation and feature encoding, position encoding, Transformer feature extraction and classification decision modules, thereby realizing the automation and parallelization of protocol recognition and parameter parsing.
This instrument enables automatic identification of communication protocols and parsing of multiple signal parameters from raw IQ data, improving analysis efficiency and accuracy, lowering the technical threshold, and supporting comprehensive wireless communication signal testing across multiple application scenarios.
Smart Images
Figure CN122053736A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of wireless communication technology, and in particular to a method and device for protocol identification and parameter parsing of wireless communication signals based on deep learning. Background Technology
[0002] Traditional communication protocol analysis methods have several limitations. They require expensive equipment such as high-end spectrum analyzers and highly specialized knowledge from the testers. Engineers must perform signal observation, protocol deduction, and parameter measurement step by step. Furthermore, over 90% of users report that signal testers require repeated parameter adjustments, a process that can take anywhere from minutes to hours. This inefficient workflow is completely unsuitable for applications requiring real-time response, such as electromagnetic monitoring and emergency communications. Therefore, there is an urgent need for highly efficient communication signal analysis instruments that can be automated, intelligent, and software-based.
[0003] With the rapid development of artificial intelligence technology, some research has attempted to introduce deep learning models into the field of communication signal analysis. However, some related research on communication signal analysis has only focused on protocol identification, failing to further analyze the key configurations of the protocol, such as the actual bandwidth, modulation and coding schemes, and guard intervals. Although some research has been able to further analyze parameters when dealing with protocols like Wi-Fi 802.11a / ac / ax, this parameter analysis is fixed. Identifying "802.11ac" itself means that it may operate in the 5GHz band and have a specific channel bonding mode, but this is not a dynamically derived, variable runtime parameter from the signal. Such analysis has limited significance for real-time spectrum analysis and network operation and maintenance. Furthermore, most research on parameter analysis focuses on estimating general parameters of basic modulation types (such as BPSK and QPSK), such as symbol rate and carrier frequency, far from achieving a "protocol-level" and systematic analysis.
[0004] As for the complete process of protocol identification and key parameter parsing, current research is theoretically mature: using a unified model architecture to handle different types of communication protocols. However, this singular design approach makes it difficult to achieve a balance between the accuracy of protocol identification and parameter parsing and computational efficiency. For example, modulation scheme identification requires attention to the constellation diagram features of the signal, while symbol rate estimation requires analysis of the signal's temporal periodicity. Forcibly integrating these tasks into a single model can lead to mutual interference in feature learning. Summary of the Invention
[0005] To address the problem of the inability to perform in-depth analysis of wireless communication signals, the primary objective of this invention is to provide a deep learning-based method for protocol identification and parameter parsing of wireless communication signals that can automatically identify communication protocols from raw IQ data and simultaneously and automatically parse multiple different signal parameters, thereby achieving complete intelligent analysis of communication signals.
[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for protocol identification and parameter parsing of wireless communication signals based on deep learning, the method comprising the following sequential steps:
[0007] (1) Draw the time-domain waveform, frequency-domain spectrum and modulation constellation diagram of the IQ data sequence and form a dataset. Divide the dataset into a training set and a test set.
[0008] (2) Construct a multimodal signal recognition model, the model including a cascaded first ViT model and a second ViT model;
[0009] (3) Input the training set into the multimodal signal recognition model, train the multimodal signal recognition model, and obtain the trained model: input the time domain waveform into the first ViT model, the first ViT model outputs the communication protocol to the second ViT model, and the second ViT model outputs the signal layer parameters of the communication protocol;
[0010] (4) Input the single IQ data sequence to be analyzed into the trained model and output the structured text analysis results.
[0011] Step (1) specifically refers to: firstly, sampling the communication data of NR, BT, C-V2X, ETC 512KHZ, NB-IOT DOWN, WIFI, TDD-LTE and LTE-FDD protocols to obtain multiple IQ data sequences for each protocol. For each IQ data sequence, drawing the corresponding time-domain waveform diagram, frequency-domain spectrum diagram and modulation constellation diagram. Selecting the required image type for each signal parameter in the modulation coding scheme, bandwidth configuration, frame structure parameters and resource allocation parameters for each communication protocol type. The time-domain waveform diagram, frequency-domain spectrum diagram and modulation constellation diagram together form a dataset.
[0012] Step (2) specifically refers to: both the first ViT model and the second ViT model adopt the ViT-B-16 model; wherein, the first ViT model includes:
[0013] The first data preprocessing module is used to perform size unification, normalization and tensor quantization on the input time-domain waveform.
[0014] The first image segmentation and feature encoding module is used to divide the input image into fixed-size image blocks and map them into a sequence of feature vectors.
[0015] The first location encoding module is used to introduce spatial location information;
[0016] The first Transformer feature extraction module is used to model the global relationship between image patches through a multi-layer self-attention mechanism to achieve deep feature extraction;
[0017] The first classification decision module is used to classify and label the scores of each category, and obtains the communication protocol type through the maximum value decision;
[0018] The second ViT model includes:
[0019] The second data preprocessing module is used to perform size unification, normalization, and tensor quantization on the input time-domain waveform, frequency-domain spectrogram, and modulation constellation diagram;
[0020] The second image segmentation and feature encoding module is used to divide the input image into fixed-size image blocks and map them into a sequence of feature vectors.
[0021] The second location encoding module is used to introduce spatial location information;
[0022] The second Transformer feature extraction module is used to model the global relationship between image patches through a multi-layer self-attention mechanism to achieve deep feature extraction.
[0023] The second classification decision module is used to classify and label the scores of each category, and obtains the parameter identification results of the communication protocol through the maximum value decision.
[0024] Step (3) specifically includes the following steps in sequence:
[0025] (3a) Separate the sampled IQ sequence into independent I and Q paths to generate the time-domain waveform of the signal;
[0026] (3b) The time-domain waveform of the signal is preprocessed, including size adjustment, normalization and tensor processing, to convert the time-domain waveform of the signal into a data format that meets the input requirements of the first ViT model and input it into the first ViT model. The first ViT model identifies the type of communication protocol and outputs one of the following protocols: NR, BT, C-V2X, ETC 512KHZ, NB-IOTDOWN, WIFI, TDD-LTE, LTE-FDD.
[0027] (3c) Based on the communication protocol output by the first ViT model, generate an image type that can capture the communication protocol parameter features and convert it into a format suitable for input to the ViT model and input it into the second ViT model. The second ViT model parses the signal layer parameter values and outputs the parameter recognition results.
[0028] (3d) Format the communication protocol category and parameter identification results, and output a text report or standardized data interface.
[0029] Another object of the present invention is to provide an electronic device comprising:
[0030] Processor; and
[0031] The memory stores computer program instructions that, when executed by the processor, cause the processor to perform the deep learning-based protocol identification and parameter parsing method for wireless communication signals as described above.
[0032] The present invention also provides a computer-readable storage medium having stored thereon computer program instructions, which, when executed by a processor, cause the processor to perform the protocol identification and parameter parsing method for deep learning-based wireless communication signals as described above.
[0033] As can be seen from the above technical solution, the beneficial effects of the present invention are as follows: First, the present invention can automatically identify communication protocols from raw IQ data, and simultaneously and automatically parse multiple different signal parameters, realizing complete intelligent analysis of communication signals. This replaces the process of multiple measurements and manual settings required in traditional instruments, greatly reducing the technical threshold and improving work efficiency. Second, the present invention, through multi-layer ViT model collaboration, allows specialized models to process specialized information, resulting in better performance and maintainability. The first ViT model identifies communication protocols by calling a protocol classification model, and the second ViT model, based on the protocol identification results, calls a dedicated model for each parameter to perform parameter analysis. The final output is a structured text report of communication protocols and parameters, which is easy for engineers to read and understand quickly, improving the efficiency and accuracy of communication signal analysis and facilitating spectrum management, network optimization, and operation and maintenance. Third, this invention has strong scalability; all models are based on the ViT-B-16 framework, and adding new communication protocol identification only requires adding the corresponding module. Furthermore, this invention supports server mode and local debugging mode, and can receive client requests via TCP network and return text reports of signal classification results. This helps in developing a "comprehensive wireless communication signal tester" applicable to multiple scenarios, achieving integrated and intelligent identification of various communication protocols and signal parameters. Attached Figure Description
[0034] Figure 1 This is a flowchart of the method of the present invention;
[0035] Figure 2 This is a flowchart of the ViT model protocol identification process in this invention;
[0036] Figure 3 A comparison chart showing the accuracy of protocol identification and parameter parsing. Detailed Implementation
[0037] like Figure 1 As shown, a method for protocol identification and parameter parsing of wireless communication signals based on deep learning is presented. This method includes the following sequential steps:
[0038] (1) Draw the time-domain waveform, frequency-domain spectrum and modulation constellation diagram of the IQ data sequence and form a dataset. Divide the dataset into a training set and a test set.
[0039] (2) Construct a multimodal signal recognition model, the model including a cascaded first ViT model and a second ViT model;
[0040] (3) Input the training set into the multimodal signal recognition model and train the multimodal signal recognition model to obtain the trained model: input the time domain waveform into the first ViT model, the first ViT model outputs the communication protocol to the second ViT model, and the second ViT model outputs the signal layer parameters of the communication protocol; the ViT model (Vision Transformer) is a visual Transformer model;
[0041] (4) Input the single IQ data sequence to be analyzed into the trained model and output the structured text analysis results.
[0042] Step (1) specifically refers to: firstly, sampling the communication data of NR, BT, C-V2X, ETC 512KHZ, NB-IOT DOWN, WIFI, TDD-LTE and LTE-FDD protocols to obtain multiple IQ data sequences for each protocol. For each IQ data sequence, drawing the corresponding time-domain waveform diagram, frequency-domain spectrum diagram and modulation constellation diagram. Selecting the required image type for each signal parameter in the modulation coding scheme, bandwidth configuration, frame structure parameters and resource allocation parameters for each communication protocol type. The time-domain waveform diagram, frequency-domain spectrum diagram and modulation constellation diagram together form a dataset.
[0043] In modern wireless communication systems, QAM modulation is a commonly used signal modulation method. The communication signal transmitted after QAM modulation can be captured and sampled by a signal sampling device to obtain a discrete digital sequence, namely the IQ data sequence. According to the basic principle of QAM modulation, this sequence is composed of alternating I-channel data and Q-channel data, where I represents the in-phase component and Q represents the quadrature component, with a phase difference of 90° between them. Therefore, IQ data is essentially a mapping expression of the communication signal sampling points in a two-dimensional rectangular coordinate system. Therefore, in step (1), the interleaved IQ data is separated into independent I-channels and Q-channels, with the I-channel as the real part and the Q-channel as the imaginary part, to obtain the IQ complex data sequence. Based on the IQ complex data sequence, the signal amplitude is calculated using the absolute value function to plot the basic time-domain amplitude waveform. A fast Fourier transform is then performed on the IQ complex data sequence to plot the basic frequency-domain amplitude spectrum. By increasing the sampling rate in the basic time-domain amplitude waveform and frequency-domain amplitude spectrum, precise time-domain amplitude waveforms and frequency-domain amplitude spectra are plotted. Finally, the IQ data sequence is plotted as points in the complex plane, where the x-coordinate represents the I-channel data and the y-coordinate represents the Q-channel data, thus plotting the modulation constellation diagram. Performing the above operations on each data sequence yields a large number of time-domain waveforms, frequency-domain spectra, and modulation constellation diagrams for IQ data sequences.
[0044] like Figure 2 As shown, step (2) specifically refers to: both the first ViT model and the second ViT model adopt the ViT-B-16 model, that is, the basic visual Transformer model based on 16×16 image patches; wherein, the first ViT model includes:
[0045] The first data preprocessing module is used to perform size unification, normalization and tensor quantization on the input time-domain waveform.
[0046] The first image segmentation and feature encoding module is used to divide the input image into fixed-size image blocks and map them into a sequence of feature vectors.
[0047] The first location encoding module is used to introduce spatial location information;
[0048] The first Transformer feature extraction module is used to model the global relationship between image patches through a multi-layer self-attention mechanism to achieve deep feature extraction;
[0049] The first classification decision module is used to classify and label the scores of each category, and obtains the communication protocol type through the maximum value decision;
[0050] The second ViT model includes:
[0051] The second data preprocessing module is used to perform size unification, normalization, and tensor quantization on the input time-domain waveform, frequency-domain spectrogram, and modulation constellation diagram;
[0052] The second image segmentation and feature encoding module is used to divide the input image into fixed-size image blocks and map them into a sequence of feature vectors.
[0053] The second location encoding module is used to introduce spatial location information;
[0054] The second Transformer feature extraction module is used to model the global relationship between image patches through a multi-layer self-attention mechanism to achieve deep feature extraction.
[0055] The second classification decision module is used to classify and label the scores of each category, and obtains the parameter identification results of the communication protocol through the maximum value decision.
[0056] Step (3) specifically includes the following steps in sequence:
[0057] (3a) Separate the sampled IQ sequence into independent I and Q paths to generate the time-domain waveform of the signal;
[0058] (3b) The time-domain waveform of the signal is preprocessed, including size adjustment, normalization and tensor processing, to convert the time-domain waveform of the signal into a data format that meets the input requirements of the first ViT model and input it into the first ViT model. The first ViT model identifies the type of communication protocol and outputs one of the following protocols: NR, BT, C-V2X, ETC 512KHZ, NB-IOTDOWN, WIFI, TDD-LTE, LTE-FDD.
[0059] (3c) Based on the communication protocol output by the first ViT model, generate an image type that can capture the communication protocol parameter features and convert it into a format suitable for input to the ViT model and input it into the second ViT model. The second ViT model parses the signal layer parameter values and outputs the parameter recognition results.
[0060] (3d) Format the communication protocol category and parameter identification results, and output a text report or standardized data interface.
[0061] When step (3b) identifies the communication protocol category as WIFI, the system analyzes the modulation and coding scheme, bandwidth, and guard interval of WIFI through multi-image fusion analysis. Specifically, it generates a modulation constellation diagram and a frequency domain spectrum diagram. The modulation constellation diagram is input into the WIFI subtype classification model, and the frequency domain spectrum diagram is input into the WIFI bandwidth classification model and guard interval classification model to parse the signal parameter set of WIFI modulation and coding strategy (MCS), bandwidth (BW), and guard interval type (GD). Similarly, the same applies to NR, BT, C-V2X, and NB-IoT DOWN.
[0062] When step (3b) identifies the communication protocol category as ETC 512KHZ, this is a dedicated simple protocol, and the system does not need further parsing after completing the first layer of identification.
[0063] When step (3b) identifies the communication protocol category as TDD-LTE, the system needs to input a frequency domain spectrogram into the bandwidth classification model to parse parameter values, and input a time domain waveform into the transmission direction classification model to determine the transmission direction. Then, it parses completely different parameter sets based on the transmission direction. When the transmission direction is downlink, the system inputs a basic time domain waveform into the code rate (CS) classification model and simultaneously inputs a precise frequency domain spectrogram into the resource block quantity (RB) classification model to parse the signal parameter sets for the TDD-LTE transmission direction: DOWN, BW, CS, and RB. When the transmission direction is uplink, the system inputs a time domain waveform into the Physical Uplink Control Channel Format (PCS) classification model and simultaneously inputs a frequency domain spectrogram into the Physical Resource Block Quantity (PRB) classification model to parse the signal parameter sets for the TDD-LTE transmission direction: UP, BW, PCS, and PRB. The same applies to LTE-FDD. Finally, a structured text report is output, which can be received from client requests via a TCP network and returns a text report of the signal classification results.
[0064] Experimental results show that, Figure 3 As shown, the accuracy of this invention is higher than that of the single-model method in tasks such as protocol identification, WIFI subtype identification (modulation and coding scheme), bandwidth identification, and guard interval identification. Taking WIFI subtype identification as an example, the accuracy of this invention reaches 91%, while that of the single model is only 73%. This fully demonstrates that model specialization can effectively avoid feature interference between different tasks, thereby obtaining higher analytical accuracy.
[0065] In summary, this invention achieves complete intelligent analysis of communication signals, replacing the process of multiple measurements and manual settings required in traditional instruments, greatly lowering the technical threshold and improving work efficiency. Through multi-layered ViT model collaboration, this invention allows specialized models to process specialized information, resulting in superior performance and better maintainability. The first ViT model identifies communication protocols by calling a protocol classification model, while the second ViT model, based on the protocol identification results, calls a dedicated model for each parameter to parse the parameters, ultimately outputting a structured text report of communication protocols and parameters. This facilitates quick reading and understanding by engineers, improving the efficiency and accuracy of communication signal analysis and simplifying spectrum management, network optimization, and maintenance. This invention is highly scalable; all models are based on the ViT-B-16 framework, and adding new communication protocol identification only requires adding the corresponding module. Furthermore, this invention supports server mode and local debugging mode, and can receive client requests via TCP network and return text reports of signal classification results. This contributes to the development of a "comprehensive wireless communication signal tester" applicable to multiple scenarios, achieving integrated and intelligent identification of various communication protocols and signal parameters.
[0066] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.
Claims
1. A method for protocol identification and parameter parsing of wireless communication signals based on deep learning, characterized in that: The method includes the following steps in sequence: (1) Draw the time-domain waveform, frequency-domain spectrum and modulation constellation diagram of the IQ data sequence and form a dataset. Divide the dataset into a training set and a test set. (2) Construct a multimodal signal recognition model, the model including a cascaded first ViT model and a second ViT model; (3) Input the training set into the multimodal signal recognition model, train the multimodal signal recognition model, and obtain the trained model: input the time domain waveform into the first ViT model, the first ViT model outputs the communication protocol to the second ViT model, and the second ViT model outputs the signal layer parameters of the communication protocol; (4) Input the single IQ data sequence to be analyzed into the trained model and output the structured text analysis results.
2. The method for protocol identification and parameter parsing of wireless communication signals based on deep learning according to claim 1, characterized in that: Step (1) specifically refers to: firstly, sampling the communication data of NR, BT, C-V2X, ETC 512KHZ, NB-IOT DOWN, WIFI, TDD-LTE and LTE-FDD protocols to obtain multiple IQ data sequences for each protocol. For each IQ data sequence, drawing the corresponding time-domain waveform diagram, frequency-domain spectrum diagram and modulation constellation diagram. Selecting the required image type for each signal parameter in the modulation coding scheme, bandwidth configuration, frame structure parameters and resource allocation parameters for each communication protocol type. The time-domain waveform diagram, frequency-domain spectrum diagram and modulation constellation diagram together form a dataset.
3. The method for protocol identification and parameter parsing of wireless communication signals based on deep learning according to claim 1, characterized in that: Step (2) specifically refers to: both the first ViT model and the second ViT model adopt the ViT-B-16 model; wherein, the first ViT model includes: The first data preprocessing module is used to perform size unification, normalization and tensor quantization on the input time-domain waveform. The first image segmentation and feature encoding module is used to divide the input image into fixed-size image blocks and map them into a sequence of feature vectors. The first location encoding module is used to introduce spatial location information; The first Transformer feature extraction module is used to model the global relationship between image patches through a multi-layer self-attention mechanism to achieve deep feature extraction; The first classification decision module is used to classify and label the scores of each category, and obtains the communication protocol type through the maximum value decision; The second ViT model includes: The second data preprocessing module is used to perform size unification, normalization, and tensor quantization on the input time-domain waveform, frequency-domain spectrogram, and modulation constellation diagram; The second image segmentation and feature encoding module is used to divide the input image into fixed-size image blocks and map them into a sequence of feature vectors. The second location encoding module is used to introduce spatial location information; The second Transformer feature extraction module is used to model the global relationship between image patches through a multi-layer self-attention mechanism to achieve deep feature extraction. The second classification decision module is used to classify and label the scores of each category, and obtains the parameter identification results of the communication protocol through the maximum value decision.
4. The method for protocol identification and parameter parsing of wireless communication signals based on deep learning according to claim 1, characterized in that: Step (3) specifically includes the following steps in sequence: (3a) Separate the sampled IQ sequence into independent I and Q paths to generate the time-domain waveform of the signal; (3b) The time-domain waveform of the signal is preprocessed, including size adjustment, normalization and tensor processing, to convert the time-domain waveform of the signal into a data format that meets the input requirements of the first ViT model and input it into the first ViT model. The first ViT model identifies the type of communication protocol and outputs one of the following protocols: NR, BT, C-V2X, ETC 512KHZ, NB-IOT DOWN, WIFI, TDD-LTE, LTE-FDD. (3c) Based on the communication protocol output by the first ViT model, generate an image type that can capture the communication protocol parameter features and convert it into a format suitable for input to the ViT model and input it into the second ViT model. The second ViT model parses the signal layer parameter values and outputs the parameter recognition results. (3d) Format the communication protocol category and parameter identification results, and output a text report or standardized data interface.
5. An electronic device, comprising: processor; as well as A memory storing computer program instructions, which, when executed by the processor, cause the processor to perform the protocol identification and parameter parsing method for deep learning-based wireless communication signals as described in any one of claims 1-4.
6. A computer-readable storage medium having stored thereon computer program instructions, which, when executed by a processor, cause the processor to perform the protocol identification and parameter parsing method for deep learning-based wireless communication signals as described in any one of claims 1-4.