Automated deployment of a wifi access point information map building method and system

Abstract

The application provides an automatic deployment WiFi access point information map construction method and system, which comprises: channel change feature analysis, analysis of energy distribution of the signal on a time axis, identification of an energy concentration area, and obtaining of an energy highest point PLCR; establishment of a correlation between the energy highest point PLCR, human body movement speed and human body position, acquisition and utilization of a model prediction framework, reverse solving of a trajectory, integration of human body movement constraints and space constraints in a simulation data set, utilization of speed constraints in the movement constraints to deduce a feasible region of a target position, calculation of feature data sets according to the simulation trajectory sequence, training of a data driven framework model according to the feature data sets, and processing to obtain a predicted trajectory; and SAGE-like iterative solving, utilization of a SAGE algorithm to iteratively optimize parameter estimation values until a preset convergence condition is met. The application solves technical problems such as large-scale deployment difficulty, dependence on specific antenna layout, system adaptability, and low modeling and positioning operation precision.

Description

Technical Field

[0001] This invention relates to the field of WiFi access point information map construction in indoor environments, and specifically to a method and system for automatically deploying WiFi access point information map construction. Background Technology

[0002] The Internet of Things (IoT) has made significant progress in recent years, greatly enhancing the intelligence and convenience of our lives. Whether in home life or industrial production, IoT is revolutionizing our daily routines and work patterns. Continuous innovation and miniaturization trends in sensor technology have made IoT devices more compact and sophisticated, easily integrating them into various devices and environments. This high degree of interconnectivity allows for real-time communication and collaborative operation between devices, significantly improving work efficiency and intelligence. Over the past decade, WiFi-based deviceless positioning technology has experienced rapid development. Because this technology does not require users to carry specific devices, it has shown great potential in multiple application scenarios, such as medical monitoring and home automation. The core idea of ​​deviceless positioning technology is to capture and analyze signals reflected by the human body by analyzing Channel State Information (CSI), thereby extracting motion-related characteristics to achieve target tracking. To gain a deeper understanding of the fundamental principles of deviceless sensing and positioning, researchers have introduced various theoretical models, including the well-known Fresnel region model and hyperbolic region model.

[0003] For example, the existing invention patent application document CN107016117A, entitled "Indoor Map Creation System Facilitating Timely Updates," describes a method that includes: a QR code identifier, a handheld terminal, and a data processing center. The QR code identifier is generated by a QR code generator using a user-side indoor map update webpage as the information source. The handheld terminal includes a QR code recognition module, an input module, a WiFi positioning module, and an information sending module. The data processing center includes a map building module, a storage module, an information receiving module, and an analysis and processing module used to compare the information stored in the storage module with the information received by the information receiving module, and for updating the indoor map in the map building module. The indoor map creation system provided by this invention facilitates timely updates to indoor maps. The existing invention patent application document CN116304915A, entitled "Contactless Action Recognition Method, System and Laboratory Device Based on WiFi," describes a method that includes: acquiring channel state information data through a WiFi device; preprocessing the received channel state information data to eliminate noise from machine operation and environmental interference; selecting subcarriers from the channel state information data of multiple subcarriers across each antenna link using a dynamic subcarrier selection algorithm; dynamically and statically judging the acquired channel state information data and segmenting the action range; extracting features from the action range; training the extracted features; and classifying the action based on the trained model. Although the aforementioned WiFi positioning technology has achieved a high level of accuracy, its operating principle mainly relies on the WiFi device to determine the relative spatial position, which requires a relatively accurate pre-setting of the device's location. This indicates that in the deployment of existing systems, manual calibration of access points is still necessary, undoubtedly increasing the difficulty of large-scale deployment. With the popularization of smart homes and the arrival of the Internet of Things era, the number of IoT devices connected to WiFi in the home environment is expected to continue to grow. Given that smart devices are typically designed to be lightweight and portable, coupled with the unpredictability of user behavior and the variability of the home environment, frequent changes in device location can lead to spatial perception deviations. If calibration requires user intervention, this increases the difficulty of use and limits the adaptability of the sensing system. Therefore, developing an automated calibration technique is crucial. Exploring a system that can automatically adapt to environmental changes and achieve higher-precision positioning without user intervention will be a significant challenge. Such an approach can not only reduce system deployment and maintenance costs but also improve system stability and user experience, laying a solid foundation for the advancement of intelligent sensing technology.

[0004] Current research on WiFi sensing technology largely focuses on enhancing its functionality and improving positioning accuracy, while research addressing practical application problems is relatively lacking. Firstly, the core task of WiFi sensing technology is to analyze signal changes caused by user activity and effectively model them. However, since device location changes are often brief and instantaneous, the captured signal change characteristics are relatively weak, making accurate modeling of these changes challenging. Secondly, the diverse shapes and structures of home smart devices lead to varied antenna layouts. Sensing systems relying on specific antenna layouts cannot meet the sensing needs of different devices, thus limiting their application. In light of these challenges, a method utilizing mobile robots to assist sensing has been proposed to enhance scene perception capabilities. While this method has made some progress in improving positioning accuracy, considering cost and broad applicability to indoor environments, this solution still faces many challenges in commercialization and practical deployment.

[0005] In summary, existing technologies suffer from technical problems such as difficulty in large-scale deployment, dependence on specific antenna layouts limiting system adaptability, and low accuracy in modeling and positioning operations. Summary of the Invention

[0006] The technical problem to be solved by this invention is: how to solve the technical problems of large-scale deployment difficulty, dependence on specific antenna layout restricting system adaptability, and low accuracy of modeling and positioning operations in the prior art.

[0007] This invention solves the above-mentioned technical problems by employing the following technical solution: A method for automatically deploying a WiFi access point information map includes:

[0008] S1. High-pass and low-pass filtering are applied to the signal to extract information related to human motion. The ratio of received signals from different antennas is calculated to perform phase shift correction on the original CSI signal. Human motion signal change data is obtained through time-frequency analysis. The STFT method is used to analyze the signal, analyze the energy distribution of the signal on the time axis, identify the energy concentration area, and obtain the energy highest point PLCR.

[0009] S2. Establish the correlation between the highest energy point PLCR, human movement speed, and human position. Obtain and utilize the model prediction framework to solve the trajectory in reverse. Incorporate human movement constraints and spatial constraints into the simulation dataset. Utilize the velocity constraint in the movement constraints to derive the feasible region of the target position. Generate a preset number of simulation trajectory sequences to cover the feasible region. Calculate the simulated PLCR sequence based on the simulation trajectory sequences to obtain the feature dataset. Train the data-driven framework model based on this dataset and process it to obtain the predicted trajectory.

[0010] S3, Fix the current parameter estimate θ(t) Let X be the location of the equipment. Calculate the conditional expectation of the latent variable Z relative to the observed data X, perform the expectation solution operation to obtain the latent variable, update the parameter estimate based on the expected value of the latent variable, alternately perform the expectation solution operation and the maximization operation, and use the SAGE algorithm to iteratively optimize the parameter estimate until the preset convergence condition is met.

[0011] This invention automatically constructs a WiFi device map. Based on the characteristic changes of CSI signals in an indoor environment, it relies on the inverse reasoning effect of collaborative sensing activities on device spatial attributes to build an inverse solution model. Leveraging the excellent feature matching performance of the spatial alternation generalized expectation maximization theory and the excellent batch processing performance of neural networks, it achieves rapid construction and automated deployment of WiFi access point information maps for indoor environments. This invention utilizes collaborative sensing activities in the indoor environment to achieve accurate device location through inverse reasoning of device spatial attributes.

[0012] In a more specific technical solution, in S1, in order to quantify the change information in the aforementioned signal, a correlation operation is performed on the PLCR point with the highest energy to quantify the human motion signal change data:

[0013]

[0014] In the formula, f D This represents the Doppler frequency shift of the signal, corresponding to the highest energy portion of the spectrum. Let λ represent the PLCR of the nth link, and λ represent the wavelength of the transmitted signal.

[0015] In a more specific technical solution, S2 includes:

[0016] S21. Based on Fresnel theory and referring to the change in reflection path length, establish the correlation between the energy peak point PLCR, human motion speed, and human position:

[0017]

[0018] In the formula, and Indicates user speed (v) x ,v y The projection coefficient on the reflection path.

[0019] S22. Obtain the model prediction framework, use the model prediction framework to perform batch processing operations, and perform reverse solving of preset batch trajectories.

[0020] S23. A method for self-generating simulation datasets, incorporating human motion constraints and spatial constraints into the simulation datasets, and using the velocity constraints in the motion constraints to derive the feasible region of the target location.

[0021] S24. Randomly set the initial position of the trajectory within the preset area:

[0022]

[0023] S25. Generate a preset number of simulation trajectory sequences according to equation (2) to cover the feasible region. Calculate the simulated PLCR sequence based on the simulation trajectory sequences and equipment locations, using it as the feature dataset D. e ;

[0024] S26. Design a neural network architecture and use the neural network structure for fitting operations. The neural network architecture includes an LSTM module. The LSTM module is used as the core layer of the neural network architecture to learn the mapping relationship between the user's position and the highest energy point PLCR.

[0025] S27. Set the highest energy point PLCR and the currently retrieved initial position. The data is input into the data-driven tracking model to predict the trajectory.

[0026] This invention utilizes the excellent data batch processing performance of neural networks to achieve fine-grained spatial retrieval. As the spatial grid becomes more refined, the computing resources consumed by the device for solving the problem increase exponentially. However, while ensuring the accuracy of the solution, it solves the problem that the computing resources of the device cannot meet the requirements when the granularity of spatial traversal is too large.

[0027] In a more specific technical solution, S231 uses the following logic to express motion constraints:

[0028]

[0029] In the formula, Δt represents the coordinates of the simulated trajectory at time t, and Δt represents the simulation sampling time interval, which is consistent with the STFT sliding window time interval. and These represent the simulated velocity and simulated acceleration, respectively.

[0030] The following logic is used to express space constraints:

[0031]

[0032] In the formula, M allow This indicates a feasible indoor area that can be flexibly adjusted when deploying in a new environment.

[0033] In a more specific technical solution, S25, the simulated trajectory sequence is used as the label dataset T. e The following logic is used to determine the label dataset T. eThe feature dataset is obtained through processing:

[0034]

[0035] In the formula, t represents time.

[0036] In a more specific technical solution, in S27, the predicted trajectory is obtained using the following logic.

[0037]

[0038] In the formula, F LSTM (·) represents the trained data-driven tracking framework model. This indicates the actual PLCR data collected.

[0039] This invention trains a Siamese neural network using a self-generated dataset, which not only improves training efficiency but also significantly enhances network performance. Furthermore, the framework exhibits good scalability, allowing it to be extended to assist in large-scale spatial retrieval.

[0040] In a more specific technical solution, S3 includes:

[0041] S31. Perform the expected solution operation, fixing the current parameter estimate θ. (t) The parameter estimate θ (t) Let the location of the device be set; calculate the conditional expectation of the latent variable Z relative to the observed data X; based on the conditional expectation, under the current observed data and current parameter estimates, take a weighted average of the corresponding latent variable values ​​to obtain the expected log-likelihood function Q. θ (Z):

[0042]

[0043] In the formula, L(θ) is the complete log-likelihood function, and θ is the parameter that we want to estimate.

[0044] S32. Perform the maximization operation; update the parameter estimates using conditional expectation.

[0045] S33. By alternately executing steps S31 and S32, the parameter estimate θ is fine-tuned using the SAGE algorithm. (t) This continues until the preset convergence condition is met.

[0046] This invention addresses the shortcomings of existing smart home solutions in the automated deployment of access point information maps. By converging and matching channel features and data-driven fine-grained spatial retrieval, it achieves high-precision and rapid construction of access point information maps.

[0047] In a more specific technical solution, in S32, the expected log-likelihood function Q... θ (Z) Solve for the maximum and minimum values ​​to find the expected log-likelihood function Q. θ (Z) The parameter θ when the extreme value is obtained:

[0048] θ t+1 =arg max θ Q θ (Z), (8).

[0049] In a more specific technical solution, in S33, the parameter estimate θ is tuned using the following logic. (t) :

[0050]

[0051] In the formula, θ (t) It is the parameter estimate for the t-th iteration.

[0052] This invention employs a novel parameter extraction method to obtain an accurate PLCR solution. PLCR is a relatively accurate channel variation feature obtainable under current technological conditions, and this invention is based on this feature to solve for access point information. Therefore, the accurate solution of the PLCR feature is crucial to the effectiveness of the algorithm.

[0053] In a more specific technical solution, the automated deployment of a WiFi access point information map construction system includes:

[0054] The channel variation feature analysis module is used to perform high-pass and low-pass filtering on the signal to extract human motion-related information; calculate the ratio of received signals from different antennas to perform phase offset correction on the original CSI signal; and obtain human motion signal variation data through time-frequency analysis, analyze the signal using the STFT method, analyze the energy distribution of the signal on the time axis, identify the energy concentration area, and obtain the energy highest point PLCR.

[0055] The data-driven framework training module is used to establish the correlation between the highest energy point PLCR, human motion speed, and human position. It acquires and uses the model prediction framework to solve the trajectory in reverse. It incorporates human motion constraints and spatial constraints into the simulation dataset. Using the velocity constraint in the motion constraints, it derives the feasible region of the target position and generates a preset number of simulation trajectory sequences to cover the feasible region. Based on the simulation trajectory sequences, it calculates the simulated PLCR sequence to obtain the feature dataset, which is used to train the data-driven framework model. The model is then processed to obtain the predicted trajectory. The data-driven framework training module is connected to the channel change feature analysis module.

[0056] The SAGE-like iterative solution module is used to fix the current parameter estimate θ. (t)Assuming the location is the equipment location; calculate the conditional expectation of the latent variable Z relative to the observed data X, perform the expectation solution operation to obtain the latent variable, update the parameter estimate value according to the expectation value of the latent variable, alternately perform the expectation solution operation and the maximization operation, and use the SAGE algorithm to iteratively optimize the parameter estimate value until the preset convergence condition is met. The SAGE-like iterative solution module is connected to the data-driven framework training module.

[0057] The present invention has the following advantages over the prior art:

[0058] This invention automatically constructs a WiFi device map. Based on the characteristic changes of CSI signals in an indoor environment, it relies on the inverse reasoning effect of collaborative sensing activities on device spatial attributes to build an inverse solution model. Leveraging the excellent feature matching performance of the spatial alternation generalized expectation maximization theory and the excellent batch processing performance of neural networks, it achieves rapid construction and automated deployment of WiFi access point information maps for indoor environments. This invention utilizes collaborative sensing activities in the indoor environment to achieve accurate device location through inverse reasoning of device spatial attributes.

[0059] This invention utilizes the excellent data batch processing performance of neural networks to achieve fine-grained spatial retrieval. As the spatial grid becomes more refined, the computing resources consumed by the device for solving the problem increase exponentially. However, while ensuring the accuracy of the solution, it solves the problem that the computing resources of the device cannot meet the requirements when the granularity of spatial traversal is too large.

[0060] This invention trains a Siamese neural network using a self-generated dataset, which not only improves training efficiency but also significantly enhances network performance. Furthermore, the framework exhibits good scalability, allowing it to be extended to assist in large-scale spatial retrieval.

[0061] This invention addresses the shortcomings of existing smart home solutions in the automated deployment of access point information maps. By converging and matching channel features and data-driven fine-grained spatial retrieval, it achieves high-precision and rapid construction of access point information maps.

[0062] This invention employs a novel parameter extraction method to obtain an accurate PLCR solution. PLCR is a relatively accurate channel variation feature obtainable under current technological conditions, and this invention is based on this feature to solve for access point information. Therefore, the accurate solution of the PLCR feature is crucial to the effectiveness of the algorithm.

[0063] This invention solves the technical problems of existing technologies, such as the difficulty of large-scale deployment, the dependence on specific antenna layout which restricts system adaptability, and the low accuracy of modeling and positioning operations. Attached Figure Description

[0064] Figure 1This is a schematic diagram of the basic steps of the automated deployment WiFi access point information map construction method according to Embodiment 1 of the present invention;

[0065] Figure 2 This is a schematic diagram of data flow processing for the automated deployment WiFi access point information map construction method of Embodiment 1 of the present invention;

[0066] Figure 3 This is a schematic diagram illustrating the feasibility analysis of automated solution for access point information map in Embodiment 1 of the present invention;

[0067] Figure 4 This is a schematic diagram of the signal feature analysis data stream in Embodiment 1 of the present invention;

[0068] Figure 5 This is a schematic diagram of the Fresnel zone model and velocity mapping of Embodiment 1 of the present invention;

[0069] Figure 6 This is a schematic diagram illustrating the specific steps of the SAGE-like iterative solution in Embodiment 1 of the present invention. Detailed Implementation

[0070] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0071] Example 1

[0072] In this embodiment, to ensure broad applicability of the experimental environment, a line-of-sight environment was selected as the test environment, and the system was built using the commercially available Intel 5300 wireless network card. To meet the basic operational requirements of the system, a configuration consisting of one transmitter and two receivers was designed. All related devices were placed at a height of 0.8 meters above the ground to simulate the typical height of home automation devices in practical applications. At the transmitter, a single antenna was used to perform data packet transmission; at the receiver, a triple antenna array was used to capture and decode CSI. Both the transmission and reception programs were implemented based on the open-source CSITool software library. To reduce interference from surrounding WiFi signals, the 5GHz frequency band was selected for signal tracking. In this prototype system, the transmitter sent data packets at a frequency of 1000 times per second. After data collection, a portable computer was used for data processing to calculate the user's movement trajectory. Matlab was used as the programming language throughout the system development process.

[0073] like Figure 1 and Figure 2 As shown, the automated deployment method for constructing a WiFi access point information map provided by this invention includes the following basic steps:

[0074] S1. Channel variation characteristic analysis;

[0075] like Figure 3 As shown, in this embodiment, existing sensing theories rely on precise analysis of channel variation characteristics for accurate location determination. Due to various factors such as environmental interference and equipment disturbances, it is difficult to directly extract usable location-related features from the raw CSI signal. This invention analyzes the wireless sensing portion, focusing on the dynamic portion caused by human movement. Specifically, the CSI signal includes a dynamic portion and a static portion. To achieve the above objectives, we have developed a feature extraction method.

[0076] like Figure 4 As shown, in this embodiment, to extract key information related to human movement from the signal, high-pass and low-pass filtering were first applied to the signal, focusing the analysis on signal changes caused by human activity. Since the original CSI signal is susceptible to phase shift noise, a cancellation technique was employed, namely, calculating the ratio of signals received by different antennas to correct errors caused by phase shift. When the signal passes through a human body, significant power changes occur due to the scattering characteristics of the human body. Time-frequency analysis techniques were used to track these signal changes caused by human movement. Specifically, the STFT method was used to analyze the signal, and the region with the highest energy concentration was identified by analyzing the energy distribution of the signal on the time axis. To quantify the aforementioned signal changes, the highest energy point PLCR obtained from the STFT analysis was correlated, and the formula is as follows:

[0077]

[0078] Among them, f D This represents the Doppler frequency shift of the signal, corresponding to the highest energy portion of the spectrum. Let λ represent the PLCR of the nth link, and λ represent the wavelength of the transmitted signal.

[0079] During wireless signal transmission and reception, the human torso significantly affects the signal propagation path, primarily through absorption and scattering. This effect manifests as a change in power distribution in the time-frequency spectrum, especially when human activity causes a Doppler shift in the signal. To accurately estimate and effectively compensate for the Doppler shift, this embodiment identifies and selects the region with the highest power peak in the time-frequency spectrum as the focus of analysis.

[0080] S2, Data-Driven Framework Training;

[0081] Existing tracking and sensing methods include: user behavior-related information and device location-related information; in this embodiment, the Fresnel zone is used as an example, see [link to documentation]. Figure 5 The PLCR is affected by the projection of the user's velocity onto the two reflection paths. As the reflection path length changes, the user can be positioned on a series of ellipses with the same focus during motion. By analyzing this process, the correlation between the three physical parameters—PLCR, user velocity, and user position—can be quantitatively established, as follows:

[0082]

[0083] in, and Indicates user speed (v) x ,v y The projection coefficient on the reflection path.

[0084] Although spatial retrieval can be theoretically achieved by relying on formula (2) and the channel features extracted in step S1, the computational resources required for trajectory inverse solving are too large under fine-grained conditions, making it impossible to deploy in a real environment. Therefore, the main purpose of this embodiment is to obtain the model prediction framework and complete the large-scale trajectory inverse solving by leveraging batch processing performance.

[0085] Wireless sensing, based on the acquisition and analysis of wireless communication signals, empowers systems to perceive and interpret various events, entities, or behaviors occurring in their surrounding environment by capturing, processing, and interpreting these signals. These signals may originate from wireless transmission tools such as WiFi routers, Bluetooth devices, and RFID tags. The research and application of wireless sensing technology are extensive, with researchers frequently using it to develop diverse high-tech solutions, such as precise wireless navigation, respiratory rhythm monitoring, and the construction of smart home systems. Through in-depth analysis and effective utilization of wireless signals, related systems can extract key information to facilitate effective monitoring and in-depth analysis of individual behavioral patterns.

[0086] WiFi wireless sensing is an emerging field in wireless environmental monitoring. In indoor environments, WiFi routers not only provide network access but also become multifunctional due to the environmental information contained in their signals. The properties of these signals, including dynamic changes in signal strength, path loss, and signal phase, provide a rich data source for monitoring environmental changes. This data can be used to identify and track people and objects, and even determine their precise locations without the need for additional monitoring hardware. A significant advantage of WiFi sensing technology is its reuse of existing network infrastructure, which reduces the need for additional hardware, lowers costs, and simplifies the deployment process. It improves system flexibility and economy by reducing reliance on dedicated sensors. Furthermore, WiFi sensing technology demonstrates broad applicability in applications such as indoor navigation and smart homes, providing an innovative solution for the design and implementation of intelligent systems.

[0087] The performance and effectiveness of neural network models largely depend on the quality of the training dataset. A high-quality dataset should have the following characteristics: (1) The dataset contains a sufficiently large amount of data to ensure that the neural network can fully learn the data distribution pattern. (2) The dataset is rich and diverse in content, which can ensure that the neural network can adapt to different scenarios and needs. (3) The dataset has a consistent annotation format to ensure that the neural network can understand the meaning of the data. However, datasets in the field of WiFi wireless sensing do not exhibit the above three characteristics. Existing open-source datasets have few samples and are only for specific scenarios, so the system will be completely unusable when the device location changes. Therefore, this invention designs a method for self-generating simulation datasets, incorporating human motion constraints and spatial constraints into the dataset.

[0088] In this embodiment, motion constraints are centrally introduced into the dataset. Specifically, for human motion in an indoor scene, the acceleration of the actual human motion will be within a certain range, and the probability of the target turning during movement will tend to stabilize over time. Based on the mathematical and physical relationship between velocity and acceleration, it can be inferred that the simulated velocity of human motion will vary within a feasible region without sudden or abnormal changes. Based on the relationship between velocity and position, we can derive the feasible region of the target position using velocity constraints. The above motion constraints can be expressed as follows:

[0089]

[0090] in, Δt represents the coordinates of the simulated trajectory at time t, and Δt represents the simulation sampling time interval, which is consistent with the STFT sliding window time interval. and These represent the simulated velocity and simulated acceleration, respectively. Since indoor environments typically do not involve strenuous activity, the magnitude of the acceleration is usually less than 2 m / s².2 The speed is typically less than 3 m / s. Under simulation conditions, we assign a probability of 0.05 to the occurrence of the turning event. α(t) and β(t) represent the angles between the velocity direction and the acceleration direction relative to the reference coordinate axis, respectively.

[0091] Access point information mapping refers to creating a map that displays the geographical location and related information of network access points. These access points may include Wi-Fi hotspots, mobile communication base stations, and access points of internet service providers. In the field of geographic information systems, such maps can help network operators, urban planners, and researchers understand network coverage, optimize network layout, improve service quality, and conduct network-related planning and analysis. In the field of wireless sensing, access point information is a crucial prior parameter for sensing algorithms.

[0092] In this embodiment, spatial constraints are added to the simulation dataset. These constraints mainly stem from the actual dimensions of the indoor space and the positions of objects, as expressed by the following formula:

[0093]

[0094] Among them, M allow This indicates a feasible indoor area that can be flexibly adjusted when deploying in a new environment.

[0095] In this embodiment, a series of initial positions of trajectories are randomly set within the area. Then, a sufficient number of simulated trajectory sequences are generated according to the above formula (2) to ensure coverage of the entire feasible region. These trajectory coordinates and device positions are used to calculate the simulated PLCR sequence, which is used as the feature dataset D. e ;

[0096] The path length change rate (PLCR) is a key parameter in wireless signal propagation, measuring how quickly the length of the propagation path changes over time during signal reflection. In complex propagation environments, such as indoors or urban areas, signal reflection leads to continuous changes in path length, posing challenges to signal localization and tracking. This rate of change not only reflects the dynamic nature of the signal propagation environment but also plays a crucial role in accurately estimating target location. In positioning systems, monitoring and analyzing the PLCR can more accurately capture the dynamic behavior of targets, thereby improving positioning accuracy. Furthermore, the PLCR is closely related to multipath effects and signal attenuation, factors that collectively influence the performance of wireless systems.

[0097] Furthermore, this embodiment takes into account that the above simulation process adopts a supervised learning method, and uses the corresponding trajectory sequence as the label dataset T.e Specifically, the training dataset can be represented as follows:

[0098]

[0099] Where t represents time. After the above processing, a training dataset can be obtained. Testing showed that this invention can generate 150,000 simulation data points within 7 seconds. Using this data as input to the training process, we can obtain the corresponding solution function represented by the neural network model. For a new scenario, model training can be completed within minutes without any additional manual labor.

[0100] After solving the dataset problem, the core part of the neural network, namely the neural network architecture, needs to be designed. Our approach is to use a theoretical model to guide the neural network, with the overall design philosophy emphasizing the theoretical model and de-emphasizing the neural network itself, allowing the neural network to only play a fitting role.

[0101] In this embodiment, considering the good predictive properties of LSTM for time series, a single LSTM layer is used as the core layer of the network. The LSTM neural network learns the mapping relationship between the user's location and the DPLCR. The PLCR actually obtained from the analysis in step S1 above, along with the currently retrieved initial location, is then used. By inputting the data into the designed data-driven tracking framework, the final predicted trajectory can be obtained. It can be expressed as follows:

[0102]

[0103] Among them, F LSTM (·) represents the trained data-driven tracking framework model. This indicates the actual PLCR data collected.

[0104] S3, SAGE-like iterative solution;

[0105] like Figure 6 As shown, in this embodiment, the mathematical expression of the SAGE-like algorithm includes the following specific steps:

[0106] S31. Perform the expected solution operation;

[0107] In this embodiment, the current parameter estimate θ is fixed. (t) Let X be the location of the equipment; calculate the conditional expectation of the latent variable Z relative to the observed data X. Specifically, the aforementioned conditional expectation reflects the weighted average of the possible values ​​of the latent variable given the observed data and the current parameter estimate.

[0108] The data formula is as follows:

[0109] Here, L(θ) is the complete log-likelihood function, which is used here only as a reference and can be replaced according to the needs of the algorithm. It depends on the observed data X and the latent variable Z, while θ is the parameter we want to estimate. The goal of this step is to find Q. θ The calculation expression for (Z).

[0110] The Spatial Alternating Generalized Expectation-Maximization (SGM) algorithm is an iterative estimation method for handling spatial data and spatial statistical models. This algorithm combines the generalized expectation-maximization framework with the optimization strategy of the alternating direction multiplier method, making it particularly suitable for solving high-dimensional data problems with complex spatial correlations. A key advantage of the Spatial Alternating SGM algorithm lies in its flexibility and scalability. It can easily adapt to various complex spatial statistical models, including but not limited to: spatial autocorrelation models, spatial error models, and spatial panel data models.

[0111] S32, Perform the maximization operation;

[0112] In this embodiment, the expected values ​​of the latent variables obtained in the expectation solution step are used to update the parameter estimates. Specifically, the expected log-likelihood function Q obtained in the expectation solution step is used. θ (Z) Solve for the maximum or minimum value, which can be done using analytical methods or numerical optimization techniques. The goal of the maximization step is to find Q. θ (Z) The parameter θ when the extreme value is obtained. The specific mathematical expression is as follows: θ t+1 =arg max θ Q θ (Z), (8)

[0113] In the actual operation of this embodiment, step S32 involves a complex optimization problem, especially when the number of model parameters and latent variables is large. Therefore, approximation methods or heuristic algorithms may be needed to find the optimal solution. By alternately executing the expectation-solving step and the maximization step, the SAGE algorithm gradually improves the parameter estimates until convergence is achieved. Each iteration aims to optimize the value of the objective estimation function, thereby obtaining a better parameter estimate. This process can be summarized as follows:

[0114] for:

[0115] Where, θ (t) It is the parameter estimate for the t-th iteration.

[0116] Based on the above description, the specific implementation process of the class-space alternating generalized expectation maximization algorithm proposed in this paper can be represented as follows: Set an upper limit θ for the number of iterations, an iteration error convergence threshold ζ, and generate a dedicated data-driven prediction model F based on the simulation dataset. LSTM (·), taking a scenario with one transmitter and two receivers as an example, the DFS sequence parsed by the two receivers. Location information to be obtained: device location set to The pseudocode for the algorithm can be represented as follows:

[0117]

[0118] The core of this invention lies in utilizing commercially available WiFi devices, enabling system deployment without additional hardware investment, thus significantly reducing the cost of manual operation and maintenance. The advantages of this invention lie in its excellent flexibility and adaptability to different home environments, enabling customized information map construction based on specific smart home scenarios to meet users' needs for personalization and high efficiency.

[0119] In summary, this invention automatically constructs a WiFi device map. Based on the characteristic changes of CSI signals in an indoor environment, it relies on the inverse reasoning effect of collaborative sensing activities on device spatial attributes to construct an inverse solution model. Leveraging the excellent feature matching performance of the spatial alternation generalized expectation maximization theory and the excellent batch processing performance of neural networks, it achieves rapid construction and automated deployment of WiFi access point information maps for indoor environments. This invention utilizes collaborative sensing activities in the indoor environment to achieve accurate device location through inverse reasoning of device spatial attributes.

[0120] This invention utilizes the excellent data batch processing performance of neural networks to achieve fine-grained spatial retrieval. As the spatial grid becomes more refined, the computing resources consumed by the device for solving the problem increase exponentially. However, while ensuring the accuracy of the solution, it solves the problem that the computing resources of the device cannot meet the requirements when the granularity of spatial traversal is too large.

[0121] This invention trains a Siamese neural network using a self-generated dataset, which not only improves training efficiency but also significantly enhances network performance. Furthermore, the framework exhibits good scalability, allowing it to be extended to assist in large-scale spatial retrieval.

[0122] This invention addresses the shortcomings of existing smart home solutions in the automated deployment of access point information maps. By converging and matching channel features and data-driven fine-grained spatial retrieval, it achieves high-precision and rapid construction of access point information maps.

[0123] This invention employs a novel parameter extraction method to obtain an accurate PLCR solution. PLCR is a relatively accurate channel variation feature obtainable under current technological conditions, and this invention is based on this feature to solve for access point information. Therefore, the accurate solution of the PLCR feature is crucial to the effectiveness of the algorithm.

[0124] This invention solves the technical problems of existing technologies, such as the difficulty of large-scale deployment, the dependence on specific antenna layout which restricts system adaptability, and the low accuracy of modeling and positioning operations.

[0125] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. 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. Such 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 automatically deploying a WiFi access point information map, characterized in that, The method includes: S1. Perform high-pass and low-pass filtering on the signal to extract human motion-related information; calculate the ratio of received signals from different antennas to perform phase shift correction on the original CSI signal; obtain human motion signal change data through time-frequency analysis, analyze the signal using the STFT method, analyze the energy distribution of the signal on the time axis, identify the energy concentration area, and obtain the energy highest point PLCR. S2. Establish the correlation between the highest energy point PLCR, human movement speed, and user's human position. Obtain and utilize the model prediction framework to solve the trajectory in reverse. Incorporate human movement constraints and spatial constraints into the simulation dataset. Utilize the speed constraint in the movement constraints to derive the feasible region of the target position. Generate a preset number of simulation trajectory sequences to cover the feasible region. Calculate the simulated PLCR sequence based on the simulation trajectory sequences to obtain the feature dataset. Train the data-driven framework model based on this dataset and process it to obtain the predicted trajectory. S3, Fix the current parameter estimation Let's assume the location is the equipment location; calculate the latent variables. Relative to observation data The expected value of the condition is obtained by performing an expectation calculation operation to obtain the latent variables. The parameter estimates are updated based on the expected values ​​of the latent variables. The expectation calculation operation and the maximization operation are executed alternately. The parameter estimates are iteratively optimized using a SAGE-like algorithm until the preset convergence condition is met.

2. The method for automatically deploying WiFi access point information map construction according to claim 1, characterized in that, In step S1, in order to quantify the change information in the aforementioned signal, a correlation operation is performed on the energy-highest point PLCR to quantify the human motion signal change data: (1) In the formula, This represents the Doppler frequency shift of the signal, corresponding to the highest energy portion of the spectrum. Indicates the first PLCR of each link This indicates the wavelength of the transmitted signal.

3. The method for automatically deploying WiFi access point information map construction according to claim 1, characterized in that, S2 includes: S21. Based on Fresnel theory and referring to the change in reflection path length, establish the correlation between the energy peak point PLCR, human movement speed, and user's position: (2) In the formula, and Indicates user speed Projection coefficients on the reflection path; S22. Obtain the model prediction framework, and use the model prediction framework to perform batch processing operations to perform reverse trajectory solving of a preset batch. S23. A method for self-generating simulation datasets, incorporating human motion constraints and spatial constraints into the simulation datasets, and using the velocity constraints in the motion constraints to derive the feasible domain of the target position; S24. Randomly set the initial position of the trajectory within the preset area: ； S25. Generate a preset number of simulation trajectory sequences according to equation (2) to cover the feasible region. Calculate a simulated PLCR sequence based on the simulation trajectory sequences and equipment locations to serve as the feature dataset. ; S26. Design a neural network architecture and use the neural network architecture to perform fitting operations, wherein the neural network architecture includes: an LSTM module; the LSTM module is used as the core layer of the neural network architecture, and the LSTM module is used to learn the mapping relationship between the user's body position and the highest energy point PLCR; S27. Set the highest energy point PLCR and the currently retrieved initial position. The data is input into the data-driven tracking model to predict the trajectory.

4. The method for constructing an automated WiFi access point information map according to claim 3, characterized in that, In step S231, the motion constraint is expressed using the following logic: (3) In the formula, Indicates the simulation trajectory at The coordinates of the current moment. This indicates the simulation sampling time interval, which is consistent with the STFT sliding window time interval. and These represent the simulated velocity and simulated acceleration, respectively. The spatial constraints are expressed using the following logic: (4) In the formula, This indicates a feasible indoor area that can be flexibly adjusted when deploying in a new environment.

5. The method for constructing an automated WiFi access point information map according to claim 3, characterized in that, In step S25, the simulated trajectory sequence is used as a label dataset. The following logic is used to determine the appropriate method based on the labeled dataset. The feature dataset is obtained through processing: (5) In the formula, Indicates time.

6. The method for automatically deploying WiFi access point information map construction according to claim 3, characterized in that, In step S27, the predicted trajectory is obtained by processing using the following logic. : (6) In the formula, This represents the trained data-driven tracking model. This indicates the actual PLCR data collected.

7. The method for automatically deploying WiFi access point information map construction according to claim 1, characterized in that, S3 includes: S31. Perform the desired solution operation and fix the current parameter estimate. The parameter estimates Set the device location; calculate the hidden variable. Relative to observation data The conditional expectation; based on the conditional expectation, under the current observed data and the current parameter estimation, the weighted average of the corresponding latent variable values ​​is used to obtain the expected log-likelihood function. : In the formula, It is a complete log-likelihood function. These are the parameters we want to estimate. S32. Perform the maximization operation; update the parameter estimate using the conditional expectation. S33. By alternately executing steps S31 and S32, the parameter estimate is tuned using a SAGE-like algorithm. This continues until the preset convergence condition is met.

8. The method for automatically deploying WiFi access point information map construction according to claim 7, characterized in that, In step S32, the expected log-likelihood function To find the maximum and minimum values, the expected log-likelihood function is obtained. Parameters for obtaining the maximum and minimum values : 。 9. The method for automatically deploying WiFi access point information map construction according to claim 7, characterized in that, In step S33, the parameter estimate is tuned using the following logic. : In the formula, It is the first The parameter estimates for the next iteration.

10. An automated deployment system for building WiFi access point information maps, characterized in that: The system includes: The channel variation feature analysis module is used to perform high-pass and low-pass filtering on the signal to extract human motion-related information; calculate the ratio of received signals from different antennas to perform phase offset correction on the original CSI signal; obtain human motion signal variation data through time-frequency analysis, analyze the signal using the STFT method, analyze the energy distribution of the signal on the time axis, identify the energy concentration area, and obtain the energy highest point PLCR; The data-driven framework training module is used to establish the correlation between the highest energy point PLCR, human movement speed, and user's human position, acquire and utilize the model prediction framework to solve the trajectory in reverse, incorporate human movement constraints and spatial constraints into the simulation dataset, use the speed constraint in the movement constraints to derive the feasible region of the target position, generate a preset number of simulation trajectory sequences to cover the feasible region, calculate the simulated PLCR sequence based on the simulation trajectory sequence to obtain the feature dataset, train the data-driven framework model, and process it to obtain the predicted trajectory. The data-driven framework training module is connected to the channel change feature analysis module. A SAGE-like iterative solution module is used to fix the current parameter estimates. Let's assume the location is the equipment location; calculate the latent variables. Relative to observation data The expected value of the condition is obtained by performing an expectation calculation operation to obtain latent variables. The parameter estimates are updated based on the expected values ​​of the latent variables. The expectation calculation operation and the maximization operation are performed alternately. The parameter estimates are iteratively optimized using a SAGE-like algorithm until the preset convergence condition is met. The SAGE-like iterative solution module is connected to the data-driven framework training module.

Images