A method, system, terminal, and storage medium for sensing signal segmentation based on a fundamental wireless signal model.

Abstract

This invention discloses a method, system, terminal, and storage medium for sensing signal segmentation in a wireless signal fundamental model. The method includes: dividing the original multi-input multi-output channel state information into a three-dimensional spatiotemporal frequency token sequence and preprocessing it; performing first-order temporal difference operations on adjacent preprocessed time-series samples to extract dynamic disturbance quantities and performing full-dimensional energy aggregation to generate a temporal energy mapping sequence; identifying token activation intervals by combining dynamic trigger thresholds; merging the token activation intervals in the temporal domain and removing transient interference segments to extract token sequences containing sensing target action or state features. This invention effectively removes background noise tokens and extracts key sensing feature sequences by performing semantic-level segmentation before the wireless signal enters the fundamental model, thereby improving the pre-training efficiency and generalization ability of the fundamental model in sensing tasks such as indoor and outdoor multi-target recognition and action recognition.

Description

Technical Field

[0001] This invention relates to the field of wireless sensing and artificial intelligence interdisciplinary technology, and in particular to a sensing signal segmentation method, system, terminal and storage medium for a wireless signal basic model. Background Technology

[0002] With the development of wireless communication technology, utilizing Channel State Information (CSI) for wireless sensing tasks such as gesture recognition and human motion recognition has become a research hotspot. Currently, wireless signal foundation models achieve multi-task generalization through large model architectures. However, raw CSI data contains a large amount of static environmental background and random noise irrelevant to the sensing task. Existing signal processing methods mostly use fixed-length sliding windows or manual annotation, making it difficult to automatically extract segments containing valid semantic actions from continuous signal streams. This results in a large number of invalid "noise tokens" being input into the foundation model during the pre-training stage, severely affecting the model's convergence speed and its generalization ability for downstream sensing tasks (such as target recognition and motion detection).

[0003] Therefore, existing technologies still have shortcomings. Summary of the Invention

[0004] To address the aforementioned shortcomings of existing technologies, this invention provides a method, system, terminal, and storage medium for perceptual signal segmentation based on a wireless signal fundamental model. The aim is to perform tokenization and segmentation of wireless signals in the spatiotemporal frequency dimensions before they enter the fundamental model, thereby extracting high-quality action segment sequences. The technical solution adopted by this invention is as follows: In a first aspect, the present invention provides a method for sensing signal segmentation based on a fundamental wireless signal model, the method comprising: The measurement data of the original multiple-input multiple-output channel state information is divided into a three-dimensional spatiotemporal frequency token sequence. Each token in the three-dimensional spatiotemporal frequency token sequence contains a fixed time length, a spatial antenna dimension, and a frequency subcarrier dimension. The three-dimensional spatiotemporal frequency token sequence is preprocessed, including amplitude extraction, multi-scale smoothing, and normalization. First-order time-domain difference operation is performed on adjacent preprocessed time-series samples to filter out static multipath reflection components in the environment and extract dynamic disturbance quantities. The dynamic disturbance quantities are then aggregated in full-dimensional energy in the spatial and frequency dimensions to generate a time-domain energy mapping sequence that characterizes the sensing intensity. Statistical analysis is performed on the time-domain energy mapping sequence. The statistical stationary median in the time-domain energy mapping sequence is used as the real-time environmental background noise benchmark. A dynamic trigger threshold is determined in combination with a preset scaling factor. The activation interval of a token whose energy intensity of the statistical stationary median of a series of consecutive tokens is continuously higher than the dynamic trigger threshold is identified. Based on the action continuity criterion, adjacent intervals in the Token activation interval with a time interval less than a time threshold are merged in the time domain, and transient interference segments with a duration less than the minimum semantic duration threshold are removed, thereby extracting the Token sequence containing the action or state features of the perceived target.

[0005] In one implementation, the three-dimensional spatiotemporal frequency token sequence is preprocessed, including: Extract the modulus of the complex values ​​of the three-dimensional spatiotemporal frequency token sequence; Multi-scale texture smoothing is achieved by applying Hanning window filters or normal smoothing operators to perform temporal convolution on three-dimensional spatiotemporal frequency token sequences. The three-dimensional spatiotemporal frequency token sequence is scaled to a specified feature range to eliminate device gain differences.

[0006] In one implementation, the time-domain energy mapping sequence is calculated as follows:

[0007] in, For a fixed time length, For spatial antenna dimensions, For frequency subcarrier dimension, express Time and space channels , subcarrier The signal amplitude at that location.

[0008] In one implementation, the dynamic trigger threshold is set in the following ways: The time-domain energy mapping sequence is smoothed by mean using a sliding window to eliminate transient spikes; The median of the entire smooth sequence is used as the environmental baseline noise level, and multiplied by a scaling factor to obtain the dynamic trigger threshold.

[0009] In one implementation, based on the action continuity criterion, adjacent intervals within the Token activation interval whose time interval is less than a time threshold are merged in the time domain, including: If the time interval between adjacent token activation intervals is less than the time threshold, the adjacent token activation intervals are determined to be the same continuous action, and the adjacent token activation intervals are merged in the time domain.

[0010] In one implementation, the time threshold is preset based on the physiological coherence of the target action and is used to logically merge short pause intervals belonging to the same semantic action. The minimum semantic duration threshold is used to distinguish between valid human action signals and non-perceptible disturbances caused by sudden deep fading or packet loss in the channel.

[0011] In one implementation, the method further includes: The extracted token sequence containing the action or state features of the perceived target is used as the structured input of the wireless signal basic model to support the joint pre-training of the wireless signal basic model for multiple sensing tasks, including target recognition, action classification and state trajectory tracking.

[0012] Secondly, embodiments of the present invention also provide a sensing signal segmentation system for a wireless signal fundamental model, wherein the system is used to implement the steps of the sensing signal segmentation method for a wireless signal fundamental model as described in any of the above technical solutions, and the system includes: The data partitioning module is used to partition the measurement data of the original multi-input multi-output channel state information into a three-dimensional spatiotemporal frequency token sequence. Each token in the three-dimensional spatiotemporal frequency token sequence contains a fixed time length, a spatial antenna dimension, and a frequency subcarrier dimension. The preprocessing module is used to preprocess the three-dimensional spatiotemporal frequency token sequence. The preprocessing includes amplitude extraction, multi-scale smoothing, and normalization. The energy sequence generation module is used to perform first-order time-domain difference operation on preprocessed adjacent time-series samples to filter out static multipath reflection components in the environment and extract dynamic disturbance quantities. The dynamic disturbance quantities are then aggregated in full-dimensional energy in the spatial and frequency dimensions to generate a time-domain energy mapping sequence that characterizes the sensing intensity. The activation interval identification module is used to perform statistical analysis on the time-domain energy mapping sequence, take the statistical stationary median in the time-domain energy mapping sequence as the real-time environmental background noise benchmark, determine the dynamic trigger threshold in combination with a preset scaling factor, and identify the token activation interval where the energy intensity of the statistical stationary median of several consecutive tokens is continuously higher than the dynamic trigger threshold. The Token sequence extraction module is used to merge adjacent intervals in the Token activation interval with a time interval less than a time threshold based on the action continuity criterion, and remove transient interference segments with a duration less than the minimum semantic duration threshold, thereby extracting a Token sequence containing the action or state features of the perceived target.

[0013] Thirdly, embodiments of the present invention also provide a terminal, wherein the terminal includes a memory, a processor, and a sensing signal segmentation program for a wireless signal basic model stored in the memory and executable on the processor. When the processor executes the sensing signal segmentation program for a wireless signal basic model, it implements the steps of the sensing signal segmentation method for a wireless signal basic model according to any of the above-mentioned schemes.

[0014] Fourthly, embodiments of the present invention also provide a computer-readable storage medium, wherein the computer-readable storage medium stores a sensing signal segmentation program for a wireless signal basic model, the sensing signal segmentation program for a wireless signal basic model implementing the steps of the sensing signal segmentation method for a wireless signal basic model as described in any of the above schemes on the computer-readable storage medium.

[0015] Beneficial Effects: Compared with existing technologies, this invention provides a sensing signal segmentation method for a fundamental wireless signal model. First, the measurement data of the original multi-input multi-output channel state information is divided into a three-dimensional spatiotemporal frequency token sequence. Each token in the three-dimensional spatiotemporal frequency token sequence contains a fixed time length, a spatial antenna dimension, and a frequency subcarrier dimension. Next, the three-dimensional spatiotemporal frequency token sequence is preprocessed, including amplitude extraction, multi-scale smoothing, and normalization. Then, a first-order time-domain difference operation is performed on adjacent preprocessed time-series samples to filter out static multipath reflection components in the environment and extract dynamic disturbances. The dynamic disturbances are then aggregated in both spatial and frequency dimensions to generate a time-domain energy mapping sequence characterizing the sensing intensity. Furthermore, statistical analysis is performed on the time-domain energy mapping sequence. The statistically stationary median in the time-domain energy mapping sequence is used as the real-time environmental background noise benchmark. A dynamic trigger threshold is determined by combining this with a preset scaling factor, identifying the token activation interval where the energy intensity of the statistically stationary median of several consecutive tokens is continuously higher than the dynamic trigger threshold. Finally, based on the action coherence criterion, adjacent intervals in the Token activation interval with a time interval less than the time threshold are merged in the time domain, and transient interference segments with a duration less than the minimum semantic duration threshold are removed, thereby extracting the Token sequence containing the perceived target action or state features.

[0016] This invention effectively removes redundant background noise tokens and extracts key perceptual feature sequences by performing semantic-level segmentation before the wireless signal enters the base model, thereby improving the pre-training efficiency and generalization ability of the base model in perception tasks such as indoor and outdoor multi-target recognition and action recognition. Attached Figure Description

[0017] Figure 1This is a flowchart of a preferred embodiment of the sensing signal segmentation method based on a wireless signal fundamental model according to the present invention.

[0018] Figure 2 This is a schematic diagram illustrating the technical route of the sensing signal segmentation method for a wireless signal basic model according to an embodiment of the present invention.

[0019] Figure 3 This is a schematic diagram of the overall architecture of the sensing signal segmentation method for a wireless signal basic model according to an embodiment of the present invention.

[0020] Figure 4 This is a diagram of the three-dimensional spatiotemporal frequency tokenization structure in the sensing signal segmentation method for a wireless signal basic model according to an embodiment of the present invention.

[0021] Figure 5 This is an example of the algorithm principle and segmentation diagram for extracting token sequences containing action or state features of the sensing target from the sensing signal segmentation method for a wireless signal basic model according to an embodiment of the present invention.

[0022] Figure 6 This is a technical framework diagram of a sensing signal segmentation system based on a wireless signal fundamental model, according to an embodiment of the present invention.

[0023] Figure 7 A schematic diagram of a terminal provided in an embodiment of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and effects of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0025] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content, operations, or steps, nor does it require execution in the described order. For example, some operations or steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0026] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0027] It should be understood that, in order to clearly describe the technical solutions of the embodiments of the present invention, the terms "first" and "second" are used in the embodiments of the present invention to distinguish identical or similar items with essentially the same function and effect. For example, "first control information" and "second control information" are only used to distinguish different control information and do not limit their order.

[0028] Those skilled in the art will understand that the words "first" and "second" do not limit the quantity or the order of execution, and that the words "first" and "second" do not necessarily imply that they are different.

[0029] It should also be understood that the term “and / or” as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0030] To address the problems of existing technologies, this embodiment provides a perceptual signal segmentation method for wireless signal foundation models. This embodiment performs semantic-level segmentation before the wireless signal enters the foundation model, effectively eliminating redundant background noise tokens and extracting key perceptual feature sequences, thereby improving the pre-training efficiency and generalization ability of the foundation model in perceptual tasks such as indoor and outdoor multi-target recognition and action recognition. In specific applications, this embodiment first divides the measurement data of the original multi-input multi-output channel state information into a three-dimensional spatiotemporal frequency token sequence. Each token in the three-dimensional spatiotemporal frequency token sequence contains a fixed time length, spatial antenna dimension, and frequency subcarrier dimension. Next, the three-dimensional spatiotemporal frequency token sequence is preprocessed, including amplitude extraction, multi-scale smoothing, and normalization. Then, a first-order temporal domain difference operation is performed on adjacent preprocessed time-series samples to filter out static multipath reflection components in the environment and extract dynamic disturbance quantities. The dynamic disturbance quantities are then subjected to full-dimensional energy aggregation in the spatial and frequency dimensions to generate a temporal energy mapping sequence representing the perceptual intensity. Next, statistical analysis is performed on the temporal energy mapping sequence. The statistically stationary median in the temporal energy mapping sequence is used as the real-time environmental background noise benchmark. Combined with a preset scaling factor, a dynamic trigger threshold is determined, and token activation intervals are identified where the energy intensity of the statistically stationary median of several consecutive tokens is continuously higher than the dynamic trigger threshold. Finally, based on the action continuity criterion, adjacent intervals in the token activation interval with a time interval less than a time threshold are merged in the temporal domain, and transient interference segments with a duration less than the minimum semantic duration threshold are removed, extracting a token sequence containing the perceived target action or state features.

[0031] The method in this embodiment can be applied to a terminal, which can be an intelligent product device such as a computer. Figure 1As shown in the figure, the sensing signal segmentation method for the basic wireless signal model in this embodiment specifically includes the following steps: Step S100: Divide the measurement data of the original multiple input multiple output channel state information into a three-dimensional spatiotemporal frequency token sequence. Each token in the three-dimensional spatiotemporal frequency token sequence contains a fixed time length, a spatial antenna dimension, and a frequency subcarrier dimension.

[0032] Combination Figure 2 and Figure 3 As shown in the figure, this embodiment first acquires the measurement data of the original multiple-input multiple-output channel state information (i.e., the original MIMO CSI measurement signal), and then divides it into dimensions required by the basic model. The three-dimensional spatiotemporal frequency token sequence. The measurement data of the original multiple-input multiple-output channel state information can be represented as... In this embodiment, non-overlapping three-dimensional token partitioning is applied to the three dimensions, with each token being a small cube, such as... Figure 4 As shown, for example, size (Time block × Antenna block × Frequency block), the specific size is a configurable hyperparameter that can be adjusted based on model capacity, data resolution and computing resources.

[0033] Step S200: Preprocess the three-dimensional spatiotemporal frequency token sequence. The preprocessing includes amplitude extraction, multi-scale smoothing, and normalization.

[0034] Next, combined Figure 2 As shown, this embodiment preprocesses the 3D spatiotemporal frequency token sequence based on a preprocessing module. In specific applications, the magnitude of the complex values ​​of the 3D spatiotemporal frequency token sequence is extracted to preserve the signal amplitude characteristics; and a Hanning window filter or a normal smoothing operator is applied to perform temporal convolution on the 3D spatiotemporal frequency token sequence to achieve multi-scale texture smoothing and eliminate environmental noise. In addition, this embodiment also scales the 3D spatiotemporal frequency token sequence to a specified feature range of [0,1] to eliminate device gain differences.

[0035] Step S300: Perform first-order time-domain difference operation on the preprocessed adjacent time-series samples to filter out static multipath reflection components in the environment and extract dynamic disturbance quantities. Perform full-dimensional energy aggregation of the dynamic disturbance quantities in the spatial and frequency dimensions to generate a time-domain energy mapping sequence that characterizes the sensing intensity.

[0036] Furthermore, such as Figure 2As shown, the core algorithm of this embodiment includes: A) temporal dimension segmentation; B) spatial and frequency energy aggregation; C) median energy smoothing and dynamic aggregation; and D) segment optimization and filtering. Specifically, this embodiment performs a first-order temporal difference operation on preprocessed adjacent time-series samples to filter out static multipath reflection components and extract dynamic disturbance quantities. The dynamic disturbance quantities are then subjected to full-dimensional energy aggregation in both spatial and frequency dimensions to generate a temporal energy mapping sequence representing the perceived intensity. The calculation method for the temporal energy mapping sequence is as follows:

[0037] in, For a fixed time length, For spatial antenna dimensions, For frequency subcarrier dimension, express Time and space channels , subcarrier The signal amplitude at that location.

[0038] The differential operation in this embodiment calculates the difference in signal amplitude between adjacent time points, using this physical means to filter out static multipath components in the environment that do not change over time. The full-dimensional energy aggregation in this embodiment performs a sum-of-squares operation on all subcarriers of the differential signal to obtain a time-domain energy mapping sequence characterizing the instantaneous motion intensity.

[0039] Step S400: Perform statistical analysis on the time-domain energy mapping sequence, use the statistical stationary median in the time-domain energy mapping sequence as the real-time environmental background noise benchmark, determine the dynamic trigger threshold in combination with a preset scaling factor, and identify the token activation interval where the energy intensity of the statistical stationary median of several consecutive tokens is continuously higher than the dynamic trigger threshold.

[0040] In this embodiment, the dynamic trigger threshold is set as follows: First, the time-domain energy mapping sequence is smoothed using a sliding window to eliminate instantaneous spikes. Then, the median of the entire smoothed sequence is used as the environmental baseline noise level, and multiplied by a scaling factor (e.g., 2.3) to obtain the dynamic trigger threshold. In this embodiment, the statistically stationary median is obtained by calculating the median of the energy sequence within the sliding window, which characterizes the robust noise floor under non-stationary conditions to avoid interference from impulsive outlier noise on the trigger threshold.

[0041] This embodiment directly suppresses static multipath by performing a first-order difference on the original CSI amplitude data along the time axis (its difference is close to 0); then, it calculates the sum of squares of the difference values ​​of all subcarriers at each time moment to obtain a differential energy sequence that characterizes the intensity of dynamic changes; finally, the time when the energy value is significantly higher than the background noise baseline (median × threshold multiple) is determined to be the time when dynamic action occurs, thereby achieving effective filtering of static multipath.

[0042] This embodiment identifies a token activation interval where the energy intensity of the statistically stable median of several consecutive tokens is continuously higher than the dynamic trigger threshold, and then marks the region where the energy intensity continuously exceeds the trigger threshold as the "activation interval".

[0043] Step S500: Based on the action continuity criterion, the adjacent intervals in the Token activation interval with a time interval less than the time threshold are merged in the time domain, and transient interference segments with a duration less than the minimum semantic duration threshold are removed, and the Token sequence containing the perceived target action or state features is extracted.

[0044] The time threshold in this embodiment is preset based on the physiological continuity of the target action and is used to logically group short pause intervals belonging to the same semantic action. Based on this, this embodiment statistically analyzes the time intervals within the token activation intervals. If the time interval between two adjacent activation intervals is less than the time threshold (e.g., 2000 sampling points), they are determined to be the same continuous action and time-domain merged. The minimum semantic duration threshold is used to distinguish between valid human action signals and non-perceptual disturbances caused by sudden deep fading or packet loss in the channel. Therefore, this embodiment can eliminate false trigger segments with excessively short durations (i.e., below the minimum semantic duration threshold, such as less than 1000 sampling points), ensuring the semantic quality of the output segments, and finally extracting the token sequence containing perceived target action or state features. Figure 5 This is a schematic diagram illustrating the algorithm principle and segmentation diagram of the token sequence containing the action or state features of the perceived target extracted by the sensing signal segmentation method for a wireless signal basic model according to an embodiment of the present invention. Furthermore, combined with... Figure 2 As shown in the illustration, the token sequence containing perceived target action or state features extracted in this embodiment serves as the structured input to the wireless signal base model, supporting its joint pre-training for multi-sensory tasks including target recognition, action classification, and state trajectory tracking. Experiments demonstrate that the method in this embodiment significantly improves the model's feature capture capability in complex scenarios.

[0045] In summary, this invention effectively eliminates redundant background noise tokens and extracts key perceptual feature sequences by performing semantic-level segmentation before the wireless signal enters the basic model, thereby improving the pre-training efficiency and generalization ability of the basic model in perceptual tasks such as indoor and outdoor multi-target recognition and action recognition.

[0046] Based on the above embodiments, the present invention also provides a sensing signal segmentation system for a wireless signal fundamental model. The system in this embodiment is used to implement the steps in the above method embodiments. Figure 6 As shown, the system includes: a data partitioning module 10, a preprocessing module 20, an energy sequence generation module 30, an activation interval identification module 40, and a token sequence extraction module 50. Specifically, the data partitioning module 10 is used to partition the measurement data of the original multi-input multi-output channel state information into a three-dimensional spatiotemporal frequency token sequence. Each token in the three-dimensional spatiotemporal frequency token sequence contains a fixed time length, a spatial antenna dimension, and a frequency subcarrier dimension. The preprocessing module 20 is used to preprocess the three-dimensional spatiotemporal frequency token sequence. The preprocessing includes amplitude extraction, multi-scale smoothing, and normalization. The energy sequence generation module 30 is used to perform first-order time-domain difference operations on adjacent preprocessed time-series samples to filter out static multipath reflection components in the environment and extract dynamic disturbance quantities. The dynamic disturbance quantities are then aggregated in full-dimensional energy in the spatial and frequency dimensions to generate a time-domain energy mapping sequence representing the sensing intensity. The activation interval identification module 40 is used to perform statistical analysis on the time-domain energy mapping sequence, using the statistically stationary median of the time-domain energy mapping sequence as the real-time environmental background noise benchmark, and determining the dynamic trigger threshold in combination with a preset scaling factor, to identify the token activation interval where the energy intensity of the statistically stationary median of several consecutive tokens is continuously higher than the dynamic trigger threshold. The token sequence extraction module 50 is used to merge adjacent intervals in the token activation interval with a time interval less than a time threshold in the time domain based on the action continuity criterion, and to remove transient interference segments with a duration less than the minimum semantic duration threshold, thereby extracting a token sequence containing the action or state characteristics of the perceived target.

[0047] The principles of each module in the sensing signal cutting system embodiment for the wireless signal basic model in this example are the same as the implementation process of each step in the above method embodiment, and will not be elaborated further here.

[0048] Based on the above embodiments, the present invention also provides a terminal, the principle block diagram of which can be as follows: Figure 7 As shown. The terminal may include one or more processors 100 ( Figure 7(Only one is shown in the image), memory 101, and computer program 102 stored in memory 101 and executable on one or more processors 100. For example, a sensing signal segmentation program for a wireless signal underlying model. When one or more processors 100 execute computer program 102, they can implement the various steps in the sensing signal segmentation method embodiment for a wireless signal underlying model. Alternatively, when one or more processors 100 execute computer program 102, they can implement the functions of various modules / units in the sensing signal segmentation apparatus embodiment for a wireless signal underlying model, without limitation herein.

[0049] In one embodiment, the processor 100 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0050] In one embodiment, memory 101 can be an internal storage unit of the terminal, such as a hard disk or RAM. Memory 101 can also be an external storage terminal of the terminal, such as a plug-in hard disk, smart media card (SM), secure digital card (SD), flash card, etc., all equipped on the terminal. Furthermore, memory 101 can include both internal and external storage units. Memory 101 is used to store computer programs and other programs and data required by the terminal. Memory 101 can also be used to temporarily store data that has been output or will be output.

[0051] Those skilled in the art will understand that Figure 7 The block diagram shown is merely a partial structural diagram related to the present invention and does not constitute a limitation on the terminal to which the present invention is applied. A specific terminal may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0052] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), direct memory bus RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0053] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for segmenting sensing signals based on a fundamental wireless signal model, characterized in that, The method includes: The measurement data of the original multiple-input multiple-output channel state information is divided into a three-dimensional spatiotemporal frequency token sequence. Each token in the three-dimensional spatiotemporal frequency token sequence contains a fixed time length, a spatial antenna dimension, and a frequency subcarrier dimension. The three-dimensional spatiotemporal frequency token sequence is preprocessed, including amplitude extraction, multi-scale smoothing, and normalization. First-order time-domain difference operation is performed on adjacent preprocessed time-series samples to filter out static multipath reflection components in the environment and extract dynamic disturbance quantities. The dynamic disturbance quantities are then aggregated in full-dimensional energy in the spatial and frequency dimensions to generate a time-domain energy mapping sequence that characterizes the sensing intensity. Statistical analysis is performed on the time-domain energy mapping sequence. The statistical stationary median in the time-domain energy mapping sequence is used as the real-time environmental background noise benchmark. A dynamic trigger threshold is determined in combination with a preset scaling factor. The activation interval of a token whose energy intensity of the statistical stationary median of a series of consecutive tokens is continuously higher than the dynamic trigger threshold is identified. Based on the action continuity criterion, adjacent intervals in the Token activation interval with a time interval less than a time threshold are merged in the time domain, and transient interference segments with a duration less than the minimum semantic duration threshold are removed, thereby extracting the Token sequence containing the action or state features of the perceived target.

2. The sensing signal segmentation method for a wireless signal fundamental model according to claim 1, characterized in that, Preprocessing of the three-dimensional spatiotemporal frequency token sequence includes: Extract the modulus of the complex values ​​of the three-dimensional spatiotemporal frequency token sequence; Multi-scale texture smoothing is achieved by applying Hanning window filters or normal smoothing operators to perform temporal convolution on three-dimensional spatiotemporal frequency token sequences. The three-dimensional spatiotemporal frequency token sequence is scaled to a specified feature range to eliminate device gain differences.

3. The sensing signal segmentation method for a wireless signal fundamental model according to claim 1, characterized in that, The time-domain energy mapping sequence is calculated as follows: in, For a fixed time length, For spatial antenna dimensions, For frequency subcarrier dimension, express Time and space channels , subcarrier The signal amplitude at that location.

4. The sensing signal segmentation method for a wireless signal fundamental model according to claim 1, characterized in that, The methods for setting the dynamic trigger threshold include: The time-domain energy mapping sequence is smoothed by mean using a sliding window to eliminate transient spikes; The median of the entire smooth sequence is used as the environmental baseline noise level, and multiplied by a scaling factor to obtain the dynamic trigger threshold.

5. The sensing signal segmentation method for a wireless signal fundamental model according to claim 1, characterized in that, Based on the action continuity criterion, adjacent intervals within the token activation interval whose time interval is less than a time threshold are merged in the time domain, including: If the time interval between adjacent token activation intervals is less than the time threshold, the adjacent token activation intervals are determined to be the same continuous action, and the adjacent token activation intervals are merged in the time domain.

6. The sensing signal segmentation method for a wireless signal fundamental model according to claim 5, characterized in that, The time threshold is preset based on the physiological continuity of the target action and is used to logically merge short pause intervals belonging to the same semantic action. The minimum semantic duration threshold is used to distinguish between valid human action signals and non-perceptible disturbances caused by sudden deep fading or packet loss in the channel.

7. The sensing signal segmentation method for a wireless signal fundamental model according to claim 1, characterized in that, The method further includes: The extracted token sequence containing the action or state features of the perceived target is used as the structured input of the wireless signal basic model to support the joint pre-training of the wireless signal basic model for multiple sensing tasks, including target recognition, action classification and state trajectory tracking.

8. A sensing signal segmentation system for a fundamental wireless signal model, characterized in that, The system is used to implement the steps of the sensing signal segmentation method for a wireless signal fundamental model as described in any one of claims 1-7, the system comprising: The data partitioning module is used to partition the measurement data of the original multi-input multi-output channel state information into a three-dimensional spatiotemporal frequency token sequence. Each token in the three-dimensional spatiotemporal frequency token sequence contains a fixed time length, a spatial antenna dimension, and a frequency subcarrier dimension. The preprocessing module is used to preprocess the three-dimensional spatiotemporal frequency token sequence. The preprocessing includes amplitude extraction, multi-scale smoothing, and normalization. The energy sequence generation module is used to perform first-order time-domain difference operation on preprocessed adjacent time-series samples to filter out static multipath reflection components in the environment and extract dynamic disturbance quantities. The dynamic disturbance quantities are then aggregated in full-dimensional energy in the spatial and frequency dimensions to generate a time-domain energy mapping sequence that characterizes the sensing intensity. The activation interval identification module is used to perform statistical analysis on the time-domain energy mapping sequence, take the statistical stationary median in the time-domain energy mapping sequence as the real-time environmental background noise benchmark, determine the dynamic trigger threshold in combination with a preset scaling factor, and identify the token activation interval where the energy intensity of the statistical stationary median of several consecutive tokens is continuously higher than the dynamic trigger threshold. The Token sequence extraction module is used to merge adjacent intervals in the Token activation interval with a time interval less than a time threshold based on the action continuity criterion, and remove transient interference segments with a duration less than the minimum semantic duration threshold, thereby extracting a Token sequence containing the action or state features of the perceived target.

9. A terminal, characterized in that, The terminal includes a memory, a processor, and a sensing signal segmentation program for a wireless signal basic model stored in the memory and executable on the processor. When the processor executes the sensing signal segmentation program for a wireless signal basic model, it implements the steps of the sensing signal segmentation method for a wireless signal basic model as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a sensing signal segmentation program for a wireless signal fundamental model, the sensing signal segmentation program for a wireless signal fundamental model implementing the steps of the sensing signal segmentation method for a wireless signal fundamental model as described in any one of claims 1-7 on the computer-readable storage medium.

Images