Physical frame structure analysis method and apparatus, system, storage medium

Abstract

The application discloses a physical frame structure analysis method and device, system and storage medium, comprising the following steps: S1, according to the frame structure, the received physical frame timing signal is divided into single sampling points and labeled; S2, using the PatchVision network, the frame structure analysis prediction of the divided and labeled physical frame timing signal is carried out. The technical scheme of the application solves the problem that the physical frame structure is difficult to be accurately analyzed.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of communication signal processing, and particularly relates to a physical frame structure analysis method and device, system and storage medium. BACKGROUND

[0002] In modern digital communication systems, timing modulation signals are widely used in wireless communication, wired network and satellite communication fields, and their analysis and prediction are the key to improving data transmission quality and efficiency. Timing modulation signals are often packaged as physical frames with specific structures. In physical layer communication, the physical frame structure is the basic unit of the data link layer, and contains key control information and data payload information. The existing frame structure analysis and prediction methods have the following problems:

[0003] The composition of the frame mainly includes a preamble and a data payload. The preamble is usually the starting part of the frame and can be used to synchronize the receiving device and indicate the modulation method used by the current frame. The header is an important part of the frame and mainly carries control information, address information, etc., used to identify the type of data frame, source address and destination address, etc. The data payload is the most important part of the frame and contains all the information that needs to be transmitted to the receiving end. Different communication protocols (such as the 802.11 standard of wireless local area network, the IEEE 802.3 standard of Ethernet, etc.) and different modulation methods (such as BPSK, QPSK, 8PSK, etc.) have different frame structures. Traditional analysis and prediction methods are difficult to adapt to complex and diverse frame designs, but the region division of different frame structures is roughly the same, which can be generally divided into header and tail redundancy information, header and actual transmitted data. The type of transmitted information is related to the modulation method, and these different types of information are called physical frame information units. As the basic composition unit of the physical frame timing signal, the physical frame information unit is an important basis for physical frame structure analysis. Therefore, accurate segmentation and identification of the physical frame information unit is a key technology for physical frame structure analysis.

[0004] Therefore, an innovative physical frame structure analysis method is needed, which can adapt to complex and diverse frame structures, improve the prediction accuracy of long timing signals, and integrate error detection and recovery mechanisms.

[0005] The accurate segmentation and recognition task of physical frame information unit can be essentially regarded as the panorama cutting task of time series signal. For the prediction task of time series signal, many scholars have carried out research from multiple aspects: document one (Shiyang Li; Xiaoyong Jin; Yao Xuan; Xiyou Zhou; Wenhu Chen; Yu-Xiang Wang; Xifeng Yan. Enhancing the Locality and Breaking the Memory Bottleneck of Transformer on Time Series Forecasting[J]. Statistics, 2019.) proposed the ConvTrans model, which combines the Transformer architecture and the convolution attention mechanism to enhance the model's attention to local context information, making the prediction more accurate and good for long sequence modeling. Document two (Haoyi Zhou, Shanghang Zhang, Jieqi Peng, Shuai Zhang, Jianxin Li, Hui Xiong, and Wancai Zhang. Informer: Beyond efficient transformer for long sequence time-series forecasting. In The Thirty Fifth AAAI Conference on Artificial Intelligence, volume 35, pp. 11106-11115, 2021.) uses a generative decoder that can predict the entire long time series at once, rather than step by step. This greatly improves the reasoning speed of long sequence prediction. Document three (Tian Zhou; Ziqing Ma; Qingsong Wen; Xue Wang; Liang Sun; Rong Jin. FEDformer: Frequency Enhanced Decomposed Transformer for Long-term Series Forecasting[J]. 2022.) introduces the FEDformer model, which introduces the frequency attention mechanism and position encoding to enhance the model's analysis and prediction ability for time series data, and performs well in long-term time series data prediction tasks.Document four (Haixu Wu; Jiehui Xu; Jianmin Wang; Mingsheng Long. Autoformer: Decomposition Transformers with Auto-Correlation for Long-Term Series Forecasting [J]. 2022.) proposes a model Autoformer for long-term data prediction analysis, which decomposes time series into trend and seasonal components, and can decompose more predictable components from complex time patterns, improving the accuracy and efficiency of long-term prediction.

[0006] Current researches are mostly focused on the analysis and prediction of time series signals such as traffic, power and weather, and still face challenges in the analysis and prediction of physical frame structure. Existing researches mainly focus on how to mix channels of multivariate time series signals, and believe that channel mixing can learn the relationship between different variables, so the model effect will be better. However, from the perspective of attention mode, channel mixing between channels with similar attention maps may make the effect better, but compared with all variables, the learned redundant information and error information, that is, the information affecting the model prediction, may be more, so blindly pursuing channel fusion may harm the attention effect. SUMMARY

[0007] The technical problem to be solved by the present application is to provide a physical frame structure analysis method and device, system and storage medium, to solve the problem that the physical frame structure is difficult to be accurately analyzed.

[0008] To achieve the above-mentioned purpose, the technical scheme adopted by the present application is as follows:

[0009] A physical frame structure analysis method, comprising:

[0010] Step S1, according to the frame structure, the received physical frame time series signal is divided into single sampling points and labeled;

[0011] Step S2, using PatchVision network to analyze and predict the frame structure of the divided and labeled physical frame time series signal.

[0012] As a preferred, in step S1, the collected physical frame time series signal is divided into different physical frame information units according to the frame structure, and the single sample point data in different information units is labeled; the sampled and labeled sequence can be represented as:

[0013] Q={q1q2...q N}

[0014] Wherein, N is the total number of sampling points, q iis the sampling value at the i th moment.

[0015] As preferred, in step S2, the PatchVision network comprises a PatchTST module and a physical frame structure prediction module, the PatchTST module comprises a Patching module, a mapping module and a Transformer module connected in sequence; wherein the Patching module cuts the obtained physical frame into a non-overlapping block sequence, the block sequence is mapped into a multi-dimensional feature space by the mapping module in the front-back order, and is input into the Transformer module to extract a feature vector.

[0016] The application further provides a physical frame structure analysis device, comprising:

[0017] The first processing unit is configured to divide and label the received physical frame time sequence signal according to the frame structure.

[0018] The second processing unit is configured to use the PatchVision network to perform frame structure analysis and prediction on the divided and labeled physical frame time sequence signal.

[0019] As preferred, the first processing unit divides the collected physical frame time sequence signal into different physical frame information units according to the frame structure, and labels the single sample point data in different information units; the sequence after sampling and labeling can be expressed as:

[0020] Q={q1q2...q N}

[0021] Wherein, N is the total number of sampling points, q i is the sampling value at the i th moment.

[0022] As preferred, the PatchVision network comprises a PatchTST module and a physical frame structure prediction module, the PatchTST module comprises a Patching module, a mapping module and a Transformer module connected in sequence; wherein the Patching module cuts the obtained physical frame into a non-overlapping block sequence, the block sequence is mapped into a multi-dimensional feature space by the mapping module in the front-back order, and is input into the Transformer module to extract a feature vector.

[0023] The application further provides a physical frame structure analysis system, comprising a memory and a processor, the memory stores a computer program run by the processor, and the computer program performs a physical frame structure analysis method when run by the processor.

[0024] The application further provides a storage medium, wherein a computer program is stored on the storage medium, and the computer program performs the physical frame structure analysis method when running.

[0025] The application adopts the PatchTST network to extract features of the sampled physical frame timing signal, and realizes frame structure analysis of the physical frame timing signal through header prediction. BRIEF DESCRIPTION OF DRAWINGS

[0026] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are only a part of the present application, and other drawings can be obtained by those skilled in the art without any creative effort on the basis of the provided drawings.

[0027] Figure 1 The physical frame structure analysis method flowchart of the present application embodiment is shown in the figure.

[0028] Figure 2 The physical frame information unit division and sampling schematic diagram is shown in the figure.

[0029] Figure 3 The implementation flowchart of the PatchTST-based physical frame timing signal panoramic prediction technology is shown in the figure.

[0030] Figure 4 The sampling and labeling schematic diagram of the training set physical frame timing signal is shown in the figure.

[0031] Figure 5 The average recognition rate schematic diagram of the physical frame timing signal structure analysis under different input conditions and models is shown in the figure.

[0032] Figure 6 The average intersection over union schematic diagram of the physical frame timing signal structure analysis under different input conditions and models is shown in the figure. DETAILED DESCRIPTION

[0033] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without any creative effort belong to the protection scope of the present application.

[0034] In order to make the above objectives, characteristics and advantages of the present application more apparent, further specific embodiments will be described in detail with reference to the accompanying drawings and specific embodiments.

[0035] Embodiment 1

[0036] As shown in the Figure 1 , the embodiment of the present application provides a physical frame structure analysis method, which comprises:

[0037] Step S1, according to the frame structure, the received physical frame time sequence signal is divided into single sampling points and labeled;

[0038] Step S2, using PatchVision network to analyze and predict the frame structure of the divided and labeled physical frame time sequence signal.

[0039] As an embodiment of the present application, in step S1, as shown in Figure 2 , the receiver divides the collected physical frame time sequence signal into different physical frame information units according to the frame structure, and labels the single sample data in different information units; the sequence after sampling and labeling can be expressed as:

[0040] Q={q1q2...q N}

[0041] Wherein, N is the total number of sampling points, q i is the sampling value at the i-th moment.

[0042] As an embodiment of the present application, as shown in Figure 3 , the PatchVision network comprises a PatchTST module and a physical frame structure prediction module; wherein the PatchTST module comprises a Patching module, a mapping module and a Transformer module connected in sequence; the Patching module cuts the obtained physical frame into a non-overlapping block sequence, the block sequence is mapped to a multi-dimensional feature space in sequence through the mapping module, and is input into the Transformer module to extract a feature vector.

[0043] Further, the Patching module divides the physical frame data into a plurality of blocks (patches), each block representing a time window, which can more effectively capture local semantic information, expressed as:

[0044]

[0045] Wherein, N p is the total number of divided blocks, P j is the j-th block, which can be expressed as:

[0046] Pj = {q j1 q j2 ...q jM}

[0047] where M is the number of sampling points in each block.

[0048] Each block sampling point is mapped to a D-dimensional feature space by the linear layer of the mapping module, so that the subsequent Transformer module can extract features, denoted as

[0049] x j = f([q j1 q j2 ...q jM ])

[0050] where, is a D-dimensional feature vector.

[0051] N p blocks constitute a feature matrix Input into the Transformer module containing H attention heads (Attention Head); the self-attention layer of the Transformer module divides the Query, Key and Value parameters into H parts, and then each part is independently processed through a separate Attention Head. Therefore, these Attention calculation results are finally fused together to form the final attention score. Through the self-attention layer, the PatchTST module can effectively identify the relationship between each part of the sequence and capture the time dependence. The attention formula can be expressed as:

[0052]

[0053] where Q, K and V represent Query (query), Key (key) and Value (value) respectively. k d represents the dimension of the attention head, usually Through feature extraction by the Transformer encoder, a physical frame d-dimensional feature map is obtained, denoted as

[0054] Further, the physical frame structure prediction module is used to convert the extracted feature map (Feature Map) into pixel-level classification prediction, as shown in Figure 4 The feature map is first input into a one-dimensional convolution model, and let N W represent the number of frame information unit categories contained in the physical frame, then in the case of considering the prediction class, there are N W +1 label categories in the panoramic prediction task. Therefore, the one-dimensional convolution formula can be expressed as:

[0055]

[0056] In order to achieve panoramic prediction, the prediction results are expanded using a linear layer, which can be expressed as:

[0057]

[0058] Where N is the total number of sampling points, Then it is the probability matrix of the label corresponding to each quantization point.

[0059] The embodiment of the present invention samples the physical frame timing signal and marks the frame information unit label corresponding to the sampling point; uses the PatchTST module to realize feature extraction of the sampling point, and adopts the physical frame structure prediction module to realize analysis of the physical frame structure.

[0060] Example:

[0061] like Figure 3 As shown, an embodiment of the present invention provides a physical frame structure analysis method based on PatchTST, including:

[0062] Step 1: Construct a training set for physical frame timing signal structure analysis

[0063] Construct a training set of physical frame timing signals, sample the physical frame timing signals uniformly, and mark the sampling points in the training set. The physical frame sampling points are marked as W i (i=1,2,...,N W ),like Figure 4 In the present invention, a total of 7 different physical frame information units are involved, and different physical frame information units have different time lengths. All header and tail redundant information in the physical frame are classified into 1 category, so a total of 8 labels are included in the panoramic prediction task.

[0064] The second step is to train the physical frame timing signal structure analysis network based on PatctTST;

[0065] Step 3: Construct a test set for physical frame timing signal structure analysis;

[0066] Step 4: Test the physical frame timing signal structure analysis network based on PatchTST.

[0067] The recognition results of the physical frame information unit when the input information of different modes is sent to the network and the recognition results of the physical frame information unit by different networks under the same input mode are as follows: Figure 5 、 6 shown.

[0068] Example 2:

[0069] The embodiment of the present application also provides a physical frame structure analysis device, comprising:

[0070] The first processing unit is used for dividing and marking the received physical frame timing signal according to the frame structure;

[0071] The second processing unit is used for performing frame structure analysis prediction on the divided and marked physical frame timing signal by using the PatchVision network.

[0072] As an embodiment of the present application, the first processing unit divides the collected physical frame timing signal into different physical frame information units according to the frame structure, and marks the single sample point data in different information units; the sequence after sampling and marking can be expressed as:

[0073] Q={q1q2...q N}

[0074] Wherein, N is the total number of sampling points, q i is the sampling value at the i-th moment.

[0075] As an embodiment of the present application, the PatchVision network comprises a PatchTST module and a physical frame structure prediction module, the PatchTST module comprises a Patching module, a mapping module and a Transformer module connected in sequence; wherein the Patching module cuts the obtained physical frame into a non-overlapping block sequence, the block sequence is mapped into a multi-dimensional feature space by the mapping module in the front-back order, and is input into the Transformer module to extract a feature vector.

[0076] Embodiment 3:

[0077] The embodiment of the present application also provides a physical frame structure analysis system, comprising a memory and a processor, the memory stores a computer program run by the processor, and the computer program performs a physical frame structure analysis method when run by the processor.

[0078] Embodiment 4:

[0079] The embodiment of the present application also provides a storage medium, the storage medium stores a computer program, and the computer program performs a physical frame structure analysis estimation method when run.

[0080] The above-described embodiments only describe the preferred modes of the present application, and do not limit the scope of the present application, and various modifications and improvements of the technical solutions of the present application made by those skilled in the art without departing from the design spirit of the present application shall fall within the protection scope of the present application determined by the claims.

Claims

1. A physical frame structure analysis method, characterized in that: include: Step S1: Divide the received physical frame timing signal into single sampling points and mark them according to the frame structure; Step S2: Use the PatchVision network to perform frame structure analysis and prediction on the divided and labeled physical frame timing signals; The PatchVision network includes a PatchTST module and a physical frame structure prediction module. The PatchTST module consists of a Patching module, a mapping module, and a Transformer module connected in sequence. The Patching module cuts the acquired physical frame into a non-overlapping block sequence, which is mapped to a multidimensional feature space in a sequential order through the mapping module and input into the Transformer module to extract feature vectors. The Patching module divides the physical frame data into multiple blocks, each of which represents a time window, expressed as: Among them, N p is the total number of partition blocks, P j is the jth block, expressed as: P j ={q j1 q j2 ...q jM } Where M is the number of sampling points in each block; Each block sampling point is mapped to the D-dimensional feature space through the linear layer of the mapping module, which is expressed as x j =f([q j1 q j2 ...q jM ]) in, is a D-dimensional feature vector; N p The feature matrix of the block The input is fed into a Transformer module containing H attention heads. The self-attention layer of the Transformer module splits the query, key, and value parameters into H parts, and then processes each part independently through a separate attention head. The attention calculation results are finally fused together to form the final attention score. The attention formula is expressed as: Among them, Q, K and V represent Query, Key and Value respectively; d k Represents the dimension of the attention head, usually Through the feature extraction of the Transformer encoder, the d-dimensional feature map of the physical frame is obtained, which is expressed as The physical frame structure prediction module is used to convert the extracted feature map into pixel-level classification prediction, specifically: feature map First input into the one-dimensional convolution model, let N W Indicates the number of frame information unit types contained in the physical frame. Considering the prediction class, the panoramic prediction task contains N W +1 label category, the one-dimensional convolution formula is expressed as: The prediction results are expanded using a linear layer and expressed as: Where N is the total number of sampling points, Then it is the probability matrix of the label corresponding to each quantization point.

2. A physical frame structure analysis device for implementing the physical frame structure analysis method according to claim 1, characterized in that: include: A first processing unit is configured to divide and mark the received physical frame timing signal into single sampling points according to the frame structure; The second processing unit is used to use the PatchVision network to perform frame structure analysis and prediction on the divided and labeled physical frame timing signals.

3. A physical frame structure analysis system, characterized in that: include: A memory and a processor, wherein the memory stores a computer program executed by the processor, and when the computer program is executed by the processor, the physical frame structure analysis method according to claim 1 is executed.

4. A storage medium, characterized in that The storage medium stores a computer program, which executes the physical frame structure analysis and estimation method according to claim 1 when running.

Images