A user presence perception method based on channel state information

Abstract

The present invention discloses a user presence perception method based on channel state information, which belongs to the field of electrical engineering and automation technology. Data filtering: the CSI acquisition end of one antenna is selected as the research object, and it is set to sample the CSI amplitude matrix H of N subcarriers at an S packet rate within T seconds. In order to avoid the negative impact of adverse factors such as measurement errors and environmental noise on personnel detection, it is necessary to perform data filtering on the time series data of each subcarrier in the CSI amplitude matrix H. In the present invention, the main advantages of the technology are: 1. High precision: the data characterization method using the modified Rayleigh distribution model provides high-precision detection results; 2. Fast speed: thanks to the collaborative computing mode and rolling computing method, the presence or absence of people can be detected within seconds; 3. Low cost: personnel detection is performed based on Wi-Fi signals, ensuring the normal use of the smart home network without introducing additional equipment.

Description

Technical Field

[0001] The present invention belongs to the field of electrical engineering and automation technology, and in particular relates to a user presence perception method based on channel state information. Background Art

[0002] In order to achieve intrusion detection, presence detection, elderly care and other needs, various devices using video surveillance, acoustic monitoring, infrared detection and other methods have been widely used in the market, but there is a disadvantage that additional equipment needs to be introduced. Due to the universal application of Wi-Fi devices, the detection method based on Wi-Fi can significantly overcome the cost increase and installation difficulties caused by the introduction of additional equipment in traditional solutions, so it has the characteristics of device-free.

[0003] As the main indicator of Wi-Fi multipath propagation results, channel state information can fully reflect the changes caused by user actions during signal propagation, and has therefore received extensive research and attention. For example, related technologies use deep learning algorithms such as support vector machines and neural networks to explore the multi-classification rules of CSI when users are in different states, and use this as an offline feature fingerprint library to compare with real-time feature fingerprints to determine whether there is a user state; related technologies use Doppler velocity motion models to measure the difference in CSI when users have different behaviors, thereby estimating the user's walking parameters, and related technologies determine whether a user exists based on the difference in correlation coefficients between subcarrier channels in different user states.

[0004] The main shortcomings of the current technology are that it generally requires a large amount of calculation and relies on dedicated network card devices, which are costly, have large delays, are not accurate enough, and are not suitable for user information perception. Specifically, on the one hand, commonly used computing devices, such as the edge computing smart gateway ECG, require a lot of computing power when solving feature fingerprints offline, thereby delaying other intelligent control processes, while the computing chip at the CSI acquisition end is idle; on the other hand, most existing technologies are based on CSI sequences within a longer time window to determine real-time personnel detection, but are affected by network delays, demodulation, and time-consuming calculations, and in practice it is difficult to achieve a judgment time scale within seconds. Summary of the invention

[0005] The purpose of the present invention is to propose a user presence perception method based on channel state information, which uses a communication module widely embedded in a smart home device network composed of gateways and terminals, combined with Wi-Fi channel state information (CSI), in a terminal and gateway collaborative calculation manner, and adopts a modified Rayleigh distribution model to perceive the user's presence information with high accuracy, speed, and low cost.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] A method for user presence perception based on channel state information specifically comprises the following steps:

[0008] S1. Data filtering: The CSI acquisition end of one antenna is selected as the research object, and it is set to sample the CSI amplitude matrix H of N subcarriers at a packet rate of S within T seconds. In order to avoid the negative impact of measurement errors and environmental noise on personnel detection, the time series data of each subcarrier in the CSI amplitude matrix H needs to be filtered;

[0009] S2, based on S1, select Hampel filtering to remove abnormal points caused by measurement errors, and then use wavelet denoising to reduce environmental noise caused by temperature and humidity changes;

[0010] S3. Modified Rayleigh distribution is used to characterize the distribution of CSI amplitude: The subcarriers of CSI do not interfere with each other because they are orthogonal to each other. Therefore, it can be assumed that they meet the wide sense stationary incoherent scattering (WSSUS) criterion, so the amplitude of the composite signal obeys the Rayleigh distribution with different parameters.

[0011] S4. Feature selection: Use the CSI amplitude characteristics to characterize the manned and unmanned status of power users, and propose the following feature quantities: shape parameter σ and right deviation correction coefficient β, subcarrier CSI amplitude change state δ.

[0012] S5. Constructing characteristic fingerprints for personnel detection: Based on the CSI amplitude characteristics, the corresponding personnel detection judgment matrix can be constructed. Affected by the frequency selectivity effect, each subcarrier has different CSI perception capabilities with or without user status, so it is necessary to screen out m sensitive subcarriers as detection objects to improve the calculation rate. At the same time, once the CSI acquisition end is installed, the position will hardly change, so it is only necessary to screen sensitive subcarriers when calculating the historical characteristic fingerprints in the first pre-preparation stage. In view of the fact that the proposed feature quantities are all related to σ, the screening method is to select the first m subcarriers in descending order of σ difference in the historical characteristic fingerprints with and without user status;

[0013] S6. Improve the calculation rate based on the rolling calculation mode: A rolling calculation mode is proposed to improve the timeliness of calculating the real-time feature fingerprint matrix. In the CSI amplitude matrix containing T data packets, this mode reduces the CSI amplitude matrix updated once for the original T data packets to k / s seconds. The value of k is a comprehensive consideration of the calculation efficiency and detection accuracy, and is generally selected as 50 data packets.

[0014] S7. The presence status of people in the room is determined by the minimum distance criterion. In order to measure the difference between the current CSI data and the data distribution under the two states, the F-norm of the feature matrix difference can be used to describe the distance between the real-time feature fingerprint and the historical feature fingerprint matrix with or without user status. The status of the one with the closest distance is taken as the final judgment result.

[0015] As a further description of the above technical solution:

[0016] In S2, it should be noted that the temperature and humidity changes are usually high-frequency noise, while the signal changes caused by user activities are mostly in the low-frequency range, so the advantage of wavelet denoising in high-frequency filtering is used to increase the detection accuracy.

[0017] As a further description of the above technical solution:

[0018] In S4, σ and β are the only two parameters describing the CSI amplitude distribution, which can accurately measure the user status. The former affects the shape of the probability density, and the latter is used to right-shift the probability distribution.

[0019] As a further description of the above technical solution:

[0020] In S4, the user state can be divided into a user static state (such as lunch break) and a user action state (such as walking), and the user static state and the non-user state have no effect on the multipath propagation of the signal. The CSI amplitude distribution of the subcarrier is in a steady state, but the user action will disturb its distribution. When the user walks or moves, the shape parameter σ changes significantly. Therefore, the subcarrier CSI amplitude change state δ is defined to characterize the user action.

[0021] As a further description of the above technical solution:

[0022] In S6, this method improves timeliness while making full use of historical data. Furthermore, the proposed mode not only does not affect the calculation of feature quantities in the absence of users, but also can more finely characterize the time domain characteristics of feature quantities in the user action state.

[0023] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0024] In the present invention, it is possible to determine whether a room is occupied or not in a gateway-terminal smart home network using Wi-Fi as the main communication method.

[0025] The main advantages of the technology are: 1. High precision: the data characterization method of the modified Rayleigh distribution model provides high-precision detection results; 2. Fast speed: thanks to the collaborative computing mode and rolling calculation method, it can detect whether there is someone or no one within seconds; 3. Low cost: personnel detection is based on Wi-Fi signals, ensuring the normal use of the smart home network without the need to introduce additional equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 A schematic diagram of a CSI-based personnel detection solution of a user presence perception method based on channel state information proposed by the present invention;

[0027] Figure 2 A schematic diagram of the topological structure of related devices of a user presence perception method based on channel state information proposed by the present invention;

[0028] Figure 3 A schematic diagram of a collaborative computing mode of ECG and CSI acquisition terminals of a user presence perception method based on channel state information proposed by the present invention;

[0029] Figure 4 A structural diagram of a real-time personnel detection process in a user presence perception method based on channel state information proposed by the present invention;

[0030] Figure 5 A schematic diagram of a parameter algorithm estimation process in a user presence perception method based on channel state information proposed by the present invention;

[0031] Figure 6 A schematic diagram of the CSI amplitude distribution of subcarriers when there is no user state in a user presence perception method based on channel state information proposed by the present invention;

[0032] Figure 7 A schematic diagram of modified Rayleigh distribution with and without user states of a user presence perception method based on channel state information proposed by the present invention;

[0033] Figure 8 A schematic diagram of a rolling calculation mode of a feature fingerprint matrix of a user presence perception method based on channel state information proposed by the present invention. DETAILED DESCRIPTION

[0034] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0035] See also Figure 1-8 The present invention provides a technical solution: a user presence perception method based on channel state information, which specifically includes the following steps:

[0036] S1. Data filtering: The CSI acquisition end of one antenna is selected as the research object, and it is set to sample the CSI amplitude matrix H of N subcarriers at a packet rate of S within T seconds. In order to avoid the negative impact of measurement errors and environmental noise on personnel detection, the time series data of each subcarrier in the CSI amplitude matrix H needs to be filtered;

[0037] The matrix H is as follows:

[0038] H=[h k,t ] N×T'

[0039] Where: h k,t is the CSI amplitude of subcarrier k at time t; T'=T*S;

[0040] S2. Based on S1, Hampel filtering is selected to eliminate abnormal points caused by measurement errors, and then wavelet denoising is used to reduce the environmental noise caused by temperature and humidity changes. In S2, it should be noted that temperature and humidity changes are usually high-frequency noise, and signal changes caused by user activities are mostly in the low-frequency range, so the advantage of wavelet denoising in high-frequency filtering is used to increase detection accuracy;

[0041] S3. Modified Rayleigh distribution is used to characterize the distribution of CSI amplitude: The subcarriers of CSI do not interfere with each other because they are orthogonal to each other. Therefore, it can be assumed that they meet the wide sense stationary incoherent scattering (WSSUS) criterion, so the amplitude of the composite signal obeys the Rayleigh distribution with different parameters.

[0042] Its probability density is as follows:

[0043]

[0044] Where: σ is the shape parameter of the Rayleigh distribution. In the standard Rayleigh distribution, all observed values ​​x i Defined as

[0045] In theory, we only need to estimate the CSI subcarrier amplitude sequence to obtain a series of CSI amplitude distribution characteristics such as the maximum probability density, mean value, and σ parameter, and then easily obtain the Rayleigh distribution parameters required to detect the user status. However, the CSI amplitude actually received by the CSI acquisition end will produce a corresponding offset, so this paper proposes a CSI modified probability density expression based on Rayleigh distribution:

[0046]

[0047] Where: β is the right-skew correction coefficient, σ is the shape parameter.

[0048] like Figure 6 As shown in Figure 2, it can be seen that the CSI amplitude distribution of the subcarrier is highly consistent with the modified Rayleigh distribution. Compared with the Gaussian kernel density estimation, the modified Rayleigh distribution can better describe the distribution form of the overall data without considering the noise.

[0049] For the above formula, we first discuss the range of x values ​​of the modified Rayleigh distribution as follows: For the standard Rayleigh distribution, x>0 for any x, and the Rayleigh distribution definition only satisfies the standard of the probability density function when x is [0, +∞). Therefore, for the modified Rayleigh distribution shown in the formula, it can be seen that it is the standard Rayleigh distribution shifted to the right by β about the x-axis. Therefore, for the modified Rayleigh distribution, in order to meet the standard of the probability density function, we should have

[0050] The maximum likelihood estimation method can be used to estimate β and σ, and the solutions are:

[0051]

[0052] As can be seen from the above formula, the expression of β involves an n+1 degree polynomial about β, which is relatively complicated to solve. Therefore, in order to adapt to the computing power of the terminal chip, a solution estimation method based on trial and error and dichotomy is adopted: by continuously adding a small value to try the zero-crossing interval of g(β) here, and then combining the dichotomy method to obtain a more accurate solution value. The algorithm flow is as follows Figure 5 As shown;

[0053] This method can quickly and accurately estimate the parameters of the modified Rayleigh distribution. Practical tests have shown that a relatively accurate solution can be obtained within five overall iterations, thus providing a basis for real-time calculations in the terminal.

[0054] S4, feature selection: Use the CSI amplitude characteristics to characterize the manned and unmanned status of power users, and propose the following feature quantities: shape parameter σ and right deviation correction coefficient β, subcarrier CSI amplitude change state δ. In S4, σ and β are the only two parameters to describe the CSI amplitude distribution, which can accurately measure the user status. The former affects the shape of the probability density, and the latter is used to right-shift the probability distribution.

[0055] In S4, the user state can be divided into the user static state (such as lunch break) and the user action state (such as walking), and the user static state and the user-free state have no effect on the multipath propagation of the signal, and the CSI amplitude distribution of the subcarrier is in a steady state, but the user action will disturb its distribution, such as Figure 7 As shown in FIG. 1 , when the user moves or acts, the shape parameter σ changes significantly. Therefore, the subcarrier CSI amplitude change state δ is defined to characterize the user action.

[0056] Given that user actions are a dynamic process, and user stillness and no user are static conditions, the values ​​of the story preparation stage are as follows:

[0057]

[0058] In the real-time detection stage, δ can be determined by determining the difference between adjacent detection periods σ:

[0059]

[0060] Where: t is the Rayleigh distribution parameter of the detection period t; Δ is the sensitive threshold of the adjacent state, and its value range will be discussed later.

[0061] S5. Constructing characteristic fingerprints for personnel detection: Based on the CSI amplitude characteristics, the corresponding personnel detection judgment matrix can be constructed. Affected by the frequency selectivity effect, each subcarrier has different CSI perception capabilities with or without user status, so it is necessary to screen out m sensitive subcarriers as detection objects to improve the calculation rate. At the same time, once the CSI acquisition end is installed, the position will hardly change, so it is only necessary to screen sensitive subcarriers when calculating the historical characteristic fingerprints in the first pre-preparation stage. In view of the fact that the proposed feature quantities are all related to σ, the screening method is to select the first m subcarriers in descending order of σ difference in the historical characteristic fingerprints with and without user status;

[0062] The characteristic fingerprint criterion is constructed based on the characteristic quantity of the sensitive subcarrier. The characteristic fingerprint matrix after normalization and dimension elimination is as follows:

[0063]

[0064] Where: CF m×3 is a characteristic fingerprint matrix composed of m sensitive subcarriers; and They are the Rayleigh distribution parameter, right deviation correction coefficient and subcarrier CSI amplitude change state of sensitive subcarrier i. At the same time, in order to ensure that each indicator has the same influence on the judgment result, the shape parameter and right deviation correction parameter need to be normalized to [0,1];

[0065] S6, improve the calculation rate based on the rolling calculation mode: propose a rolling calculation mode to improve the timeliness of the calculation of the real-time feature fingerprint matrix. In the CSI amplitude matrix containing T data packets, this mode reduces the CSI amplitude matrix updated once for the original T data packets to k / s seconds. The value of k is a value that comprehensively considers the calculation efficiency and detection accuracy. It is generally selected as 50 data packets. In S6, this method improves the timeliness while making full use of historical data. Furthermore, the proposed mode will not only not affect the calculation of feature quantities in the absence of users, but also can more finely characterize the time domain characteristics of feature quantities in the user action state;

[0066] When calculating the feature fingerprint matrix in the real-time detection stage, the CSI amplitude distribution characteristics of the subcarriers must be considered. Accumulating a certain amount of time-series CSI amplitude sequences is a prerequisite for calculating the CSI amplitude distribution. Therefore, in most studies, a certain amount of time is required to collect data to achieve this amount of accumulated data. In addition, most low-cost CSI acquisition terminals have a low packet rate S (such as 50 package / s), which seriously affects the timeliness of personnel detection. To this end, a rolling calculation mode is proposed to improve the timeliness of calculating the real-time feature fingerprint matrix.

[0067] S7. Determine the presence status of people in the room by using the minimum distance criterion. To measure the difference between the current CSI data and the data distribution under the two states, the distance between the real-time feature fingerprint and the historical feature fingerprint matrix with or without user status can be described by the F-norm of the feature matrix difference. The state of the one with the closest distance is taken as the final judgment result:

[0068]

[0069] Where: state is the real-time result of personnel detection, 1 means there is someone, 0 means there is no one; CF real CF 1 and CF 0 are real-time feature fingerprints, feature fingerprints with and without user status history; ||*|| is the F norm.

[0070] In this embodiment: the basic framework of the CSI-based detection method is described here.

[0071] CSI can reflect the channel properties of Wi-Fi multipath propagation under the IEEE802.11a / g / n protocol. It decomposes the channel measurement into a series of orthogonal subcarriers and estimates the frequency response of each subcarrier to obtain a more fine-grained channel description in the frequency domain.

[0072]

[0073] Where: H k(t) is the CSI sampling value on the kth subcarrier at time t, k≤N, N is the total number of subcarriers;

[0074] This allows the analysis of Wi-Fi signal propagation paths and their processes from a more granular perspective. Based on this, the presence detection framework is as follows: Figure 1 shown.

[0075] The technology proposed in this solution is implemented in common smart home hardware systems, and its hardware devices mainly include smart gateways, sockets, infrared and other common smart terminals.

[0076] Each terminal uses a low-cost ESP32-C3 chip, which has good storage performance and computing power. It can be integrated into terminals such as smart sockets in the smart power network as an MCU, communication module and CSI acquisition device to achieve multi-functional reuse and expansion. The overall system architecture is described as follows:

[0077] 1. CSI-related equipment topology

[0078] The topology of the personnel detection system including CSI acquisition terminal and ECG etc. related to personnel detection is as follows Figure 2 shown.

[0079] In the system, the CSI acquisition PCB board is embedded in the smart switch, smart socket or air conditioning control panel, and installed together with these devices in various power usage scenarios, and interacts with ECG through the existing WiFi module of the smart device. Figure 2 It can be seen that this type of topology reuses the Wi-Fi communication module of the perception layer, improving economic efficiency.

[0080] 2. Information interaction and computing mechanism of EGC-terminal collaboration

[0081] In order to utilize the remaining computing power of the CSI acquisition end, this solution designs a computing task allocation strategy for the ECG and CSI acquisition ends.

[0082] The collaborative computing mode of ECG and CSI acquisition ends focuses on localization of CSI acquisition ends to realize real-time detection of personnel. CSI acquisition ends realize real-time detection based on the judgment threshold trained by ECG, and then send the detection results to ECG, such as Figure 3 shown.

[0083] 3. Real-time personnel detection process based on CSI

[0084] The measured CSI value is easily affected by changes in exogenous factors such as ambient temperature and humidity, and the layout of indoor objects. This requires ECG to update the criterion threshold calculation value at a fixed frequency and transmit it to the CSI acquisition end in a timely manner to maintain the detection accuracy at a high benchmark. The criterion threshold calculation requires a certain amount of hardware performance such as computing power and memory. In engineering, the criterion threshold is often updated at a half-day frequency. ECG's threshold update process mostly requests the CSI acquisition end to send CSI data without user status in the early morning and at night, and requests the CSI acquisition end to send CSI data with user status in the morning and afternoon. ECG then calculates the criterion threshold based on these data sets.

[0085] In the collaborative computing mode of ECG and CSI acquisition, the real-time detection of user status is divided into the pre-preparation stage and the real-time detection stage, such as Figure 4 As shown. In the preparatory stage, ECG will request the CSI acquisition end to send a signal during the above-mentioned specific period, and generate a historical feature fingerprint matrix with or without user status through the proposed multi-indicator feature fingerprint algorithm. In the implementation detection stage, the CSI acquisition end needs to solve the F norm between the real-time feature fingerprint matrix and the historical feature fingerprint matrix with or without user status, respectively, to measure the difference between the real-time user status and the status with or without user status, and finally take the user status with the smaller difference value as the detection result.

[0086] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A user presence perception method based on channel state information, characterized in that: The specific steps include: S1. Data filtering: The CSI acquisition end of one antenna is selected as the research object, and it is set to sample the CSI amplitude matrix H of N subcarriers at a packet rate of S within T seconds. In order to avoid the negative impact of measurement errors and environmental noise on personnel detection, the time series data of each subcarrier in the CSI amplitude matrix H needs to be filtered; S2, based on S1, select Hampel filtering to remove abnormal points caused by measurement errors, and then use wavelet denoising to reduce environmental noise caused by temperature and humidity changes; S3. Modified Rayleigh distribution is used to characterize the distribution of CSI amplitude: The subcarriers of CSI do not interfere with each other because they are orthogonal to each other. Therefore, it can be assumed that they meet the wide sense stationary incoherent scattering (WSSUS) criterion, so the amplitude of the composite signal obeys the Rayleigh distribution with different parameters. S4, feature selection: Use the CSI amplitude characteristics to characterize the manned or unmanned state of the power user, and propose the following feature quantities: shape parameter σ and right deviation correction coefficient β, subcarrier CSI amplitude change state δ; S5. Constructing characteristic fingerprints for personnel detection: Based on the CSI amplitude characteristics, the corresponding personnel detection judgment matrix can be constructed. Affected by the frequency selectivity effect, each subcarrier has different CSI perception capabilities with or without user status, so it is necessary to screen out m sensitive subcarriers as detection objects to improve the calculation rate. At the same time, once the CSI acquisition end is installed, the position will hardly change, so it is only necessary to screen sensitive subcarriers when calculating the historical characteristic fingerprints in the first pre-preparation stage. In view of the fact that the proposed feature quantities are all related to σ, the screening method is to select the first m subcarriers in descending order of σ difference in the historical characteristic fingerprints with and without user status; S6. Improve the calculation rate based on the rolling calculation mode: A rolling calculation mode is proposed to improve the timeliness of calculating the real-time feature fingerprint matrix. In the CSI amplitude matrix containing T data packets, this mode reduces the CSI amplitude matrix updated once for the original T data packets to k / s seconds. The value of k is a comprehensive consideration of the calculation efficiency and detection accuracy, and is generally selected as 50 data packets. S7. The presence status of people in the room is determined by the minimum distance criterion. In order to measure the difference between the current CSI data and the data distribution under the two states, the F-norm of the feature matrix difference can be used to describe the distance between the real-time feature fingerprint and the historical feature fingerprint matrix with or without user status. The status of the one with the closest distance is taken as the final judgment result.

2. The method for user presence perception based on channel state information according to claim 1, characterized in that: In S2, the temperature and humidity changes are usually high-frequency noise, while the signal changes caused by user activities are mostly in the low-frequency range, so the advantage of wavelet denoising in high-frequency filtering is used to increase the detection accuracy.

3. The user presence perception method based on channel state information according to claim 1, characterized in that: In S4, σ and β are the only two parameters describing the CSI amplitude distribution, which can accurately measure the user status. The former affects the shape of the probability density, and the latter is used to right-shift the probability distribution.

4. The user presence perception method based on channel state information according to claim 1, characterized in that: In S4, the user state can be divided into a user static state and a user action state, and the user static state and the non-user state have no effect on the multipath propagation of the signal. The CSI amplitude distribution of the subcarrier is in a steady state, but the user action will disturb its distribution. When the user walks or moves, the shape parameter σ changes significantly. Therefore, the subcarrier CSI amplitude change state δ is defined to characterize the user action situation.

5. The user presence perception method based on channel state information according to claim 1, characterized in that: In S6, this method improves timeliness while making full use of historical data. Furthermore, the proposed mode not only does not affect the calculation of feature quantities in the absence of users, but also can more finely characterize the time domain characteristics of feature quantities in the user action state.

Images