CN115299917A

CN115299917A - Micro-behavior sensing method based on WiFi signal

Info

Publication number: CN115299917A
Application number: CN202210652621.2A
Authority: CN
Inventors: 古敏; 单长宁; 孙晋; 位玮; 郭欣青
Original assignee: Shandong Normal University
Current assignee: Shandong Normal University
Priority date: 2022-06-10
Filing date: 2022-06-10
Publication date: 2022-11-08

Abstract

The invention relates to a method for detecting a tiny behavior by utilizing WiFi. The micro-behavior involved in the invention refers to the behavior with the variation amplitude lower than 10 cm. The detection of the micro-behavior comprises the following steps of (1) modeling the micro-behavior: researching a Fresnel region propagation theory, and establishing a tiny behavior model on the basis of the Fresnel region theory; (2) data preprocessing: reading CSI data and extracting a CSI matrix. Then preprocessing is carried out to obtain regular tiny behavior signals; (3) subcarrier selection: due to the difference of frequency, the sensitivity of different subcarriers to the change of the tiny behavior is different, the different subcarriers have the difference of amplitude, the variance of each subcarrier is calculated, and the variance is used for quantifying the estimation of the tiny behavior of the amplitude (4) of the subcarrier: through the steps, the CSI data are filtered, and the subcarrier sensitive to the micro motion is selected, so that a relatively regular micro-behavior waveform can be obtained, and the micro-behavior can be estimated through the waveform.

Description

Micro-behavior sensing method based on WiFi signal

Technical Field

The invention belongs to the field of WiFi detection, and particularly relates to a method for sensing micro behaviors through a WiFi data building model.

Background

The WLAN basic technology: in the common technical standard protocol of IEEE802.11, the main operating frequency bands are 2.4GHz and 5GHz. The 2.4GHz center frequency is divided into 14 frequency bands, when the wireless electromagnetic wave fluctuates greatly, the receiver can find the fluctuation position and can obtain the initial signal through the relevant processing. The 5GHz signal has large energy and strong penetration capability, and the frequency band frequency is higher, so that more energy loss is caused in the transmission process. Therefore, the research based on WiFi signal perception adopts the 5GHz frequency band to carry out signal acquisition. The WLAN is a commonly used data transmission system, and an orthogonal frequency division multiplexing technology and a multiple input and output technology are used in an 802.11n standard protocol, which are key technologies of the WLAN, and the combination of the WLAN and the MAC layer and the PHY layer can improve the quality of wireless communication. Orthogonal Frequency Division Multiplexing (OFDM) is a multi-carrier transmission system compared to single carrier. The problems of intersymbol interference and the like can be effectively overcome. And after OFDM signal conversion and modulation, converting a time domain into a frequency domain through fast Fourier transform to obtain a CSI signal. The research on the OFDM technology can reduce the interference in the bandwidth and increase the sensitivity to the environment, and has important significance for researching CSI signals. The multiple-Input multiple-Output (MIMO) principle is to use multiple transmitting antennas and multiple receiving antennas to achieve the requirement of capacity expansion. MIMO technology can improve the transmission rate and reliability of a channel, and is advantageous in terms of communication capacity and coverage. The use of multiple antennas for data transmission allows for increased channel capacity without additional transmission power, thereby reducing processing complexity.

In early methods for monitoring human micro-behaviors based on WiFi signals, feature extraction was generally performed using received signal strength RSS. RSS is to directly sample a WiFi physical signal to obtain original information of the unmodulated signal, such as amplitude, phase, etc. The transmitted signal propagates along multiple paths to the receiver, and RSS is the superposition of different path components, representing the received power (dB) of the receiver:

RSS＝log ₂ (‖V‖ ² )

where V denotes a signal voltage obtained by the phase and amplitude of the multipath component. The signal strength is generally negative and is affected by three factors, namely occlusion, multipath effect and path fading, wherein the spatial and temporal changes caused by the multipath effect greatly limit the perception capability of the RSS. Even if the signal intensity is detected on a fixed path, the signal intensity can fluctuate greatly, and the perception task of monitoring the granularity as the micro-behavior characteristics of the human body cannot be completed. The RSS information can be applied to the fields of indoor positioning, target object movement detection and the like. Compared with RSS (received signal strength), the Channel State Information (CSI) belongs to a wireless channel technology with finer granularity and higher sensitivity, and describes a propagation mode from a transmitting end to a receiving end on each subcarrier. The CSI adopts OFDM technology, so that the multipath effect can be reduced and the transmission efficiency can be improved. Each antenna has a corresponding CSI value corresponding to each subcarrier, fine-grained CSI can provide phase and amplitude information for a plurality of OFDM subcarriers, and different subcarriers have different sensitivities to the perception of environmental changes. For example, the transmitting end has R antennas, the receiving end has T antennas, the number of subcarriers is M, and each data acquisition packet can analyze an R × T × M CSI matrix, which represents complete channel state information of the current transmission path. The CSI may indicate changes such as time delay, phase offset, and amplitude attenuation generated during transmission. If the sending end signal is x and the receiving end signal is y, then there are

y＝Hx+n

Where n is a noise vector, H is a channel matrix, and the noise can be represented as n-cN (0,S), it can be estimated that the Channel State Information (CSI) is an estimate of the H channel matrix. For one subcarrier, the form of CSI is as follows:

h＝he ^jsinθ

where | h | and θ represent the amplitude and phase of the CSI, respectively. The CSI can obtain information for each subcarrier of each antenna, and multipath effects caused by minor behaviors have different effects on different subcarriers, and the amplitudes of the subcarriers have significant differences. And the CSI has a higher amplitude resolution so that more subtle changes in the channel can be perceived.

The Fresnel (Fresnel) zone propagation model was used for the study of phenomena such as interference and diffraction of light in the early 19 th century, and later researchers used for the modeling of propagation of wireless signals. The Fresnel area is a concentric elliptical area which takes the receiving end of the transmitting end as a focal point. Assuming that T is the position of the transmitting end, R is the position of the receiving end, and the signal wavelength is λ, the concentric ellipses of the fresnel region include n ellipses satisfying the following formula:

|P ₁ Q _n |+|P ₂ Q _n |-|P ₁ P ₂ |＝nλ/n

wherein the point on the nth concentric ellipse is Q _n The innermost ellipse is referred to as the first fresnel region, the space between the first ellipse and the second ellipse is referred to as the second fresnel region, also referred to as the first fresnel region boundary, and so on, as shown:

the distance between the points T and R is R, the distances between the points O are R1 and R2 respectively, and the radius F of the nth Fresnel region _n Comprises the following steps:

when an object passes the boundaries of the series of concentric elliptical zones, a signal in the form of a continuous sine wave is received, which alternately produces peaks and troughs due to different phase effects caused by the object being at the boundaries of different fresnel zones. When a human body or an object has a small behavior in the fresnel region, the length of the reflection path changes with the fluctuation of the human body or the object. Since the fluctuation of a human body or an object that performs a microscopic action may be periodic, the change in amplitude may also be periodic.

The acquired CSI information usually contains a lot of noise, and it is necessary to learn and study signal data processing technology, and perform data preprocessing on the original signal, where the data preprocessing usually adopts methods such as outlier removal, interpolation, and filtering. In the research, hample filtering is adopted to remove outliers, and wavelet transform filtering is adopted to filter noise.

During signal acquisition, the original CSI value is interfered by noise in an acquisition environment, and the sudden change of the amplitude can be seen from the original CSI amplitude image. There are significant abrupt changes in some or all of the subcarriers in the data, which are clearly not due to the micro-behavior of the human body or object, and the outlier refers to a point deviating from the main signal group in the CSI amplitude. The Hampel filtering is a common method for removing abnormal points, and the experiment adopts the Hampel filtering to remove the burr points, so that the influence of sudden noise interference in the environment on signals is reduced. The principle of the Hampel filter is to identify the signals which are not in the interval [ u-r Xo, u + r Xo ] as abnormal points, obtain the signal interval fluctuating up and down by using the mean value u and the absolute value deviation r, and filter the signals which are not in the interval range as outliers. Where r depends on the environment, the most common value is 3.

After outlier removal and interpolation, the noise in the CSI needs to be removed. It is not feasible to apply a conventional time-domain or frequency-domain analysis filter to remove the high-frequency noise. Such conventional filters not only remove noise but also blur rising or falling edges in the CSI signal, which is important for small behavior feature monitoring. In the conventional low-frequency filtering methods, such as chebyshev filter, butterworth filter, etc., the experiment shows that the filtering effect is optimal for the wavelet transform.

Wavelet transform is an analysis method of time domain and frequency domain, and can obtain the characteristics of signals in both time domain and frequency domain. The wavelet transformation can change the shape of a window, analyze the local characteristics of signals, is suitable for extracting non-stationary signals, and is favorable for extracting micro-behavior signals. In order to accurately filter the CSI signal, it is necessary to determine the number of suitable decomposition layers and the wavelet basis. The number of decomposition layers has an influence on the maximum degree of decomposition of the signal and noise. Through the research on the wavelet decomposition frequency range and the sampling frequency, a large amount of noise is reserved when the number of decomposition layers is 3, and certain tiny behavior periods are filtered when the number of decomposition layers is 8.

The number of decomposition layers selected in the experiment is 5, so that a large amount of noise is filtered and certain details are reserved. Such as the original image and the filter pairs of different decomposition levels.

The result of wavelet transformation changes due to different selected wavelet bases, the selection of the wavelet bases needs to consider the conditions of support length, regularity, symmetry, similarity and the like, and different wavelet bases have different denoising effects on different types of signals. The commonly used one-dimensional denoising uses two wavelet bases of db wavelet system and sym wavelet system. The following is a comparison of db3 wavelet basis and sym8 wavelet basis at the decomposition layer number of 5 layers. It can be seen that after wavelet transform is performed by using the sym8 wavelet basis, the reserved amplitude is larger, and the extraction of the tiny behavior frequency is more facilitated.

Disclosure of Invention

The invention aims to solve the problem that micro-behaviors are difficult to identify. A WiFi-based micro-behavior awareness technique is presented.

The technology comprises four steps:

micro-behavior modeling: the micro behavior signal needs to be extracted, and in order to accurately extract the micro behavior signal and obtain the respiratory frequency, a human body or an object needs to be modeled.

Data preprocessing: firstly, reading CSI data, extracting a CSI matrix, and then drawing original data of 30 subcarriers corresponding to each antenna; after outliers are removed, some sampling points are filtered, and linear interpolation needs to be carried out on the filtered signals to ensure the integrity of sampling; after removing the abnormal value, the noise is filtered by adopting wavelet transform.

And (3) subcarrier selection: due to the difference in frequency, different subcarriers have different sensitivities to small changes in behavior. Different central frequencies of different subcarriers cause different wavelengths and also cause different multipath benefits and different shadow effects, so that the amplitudes of different subcarriers have different values. It is necessary to select a suitable subcarrier.

Estimation of the frequency of the minor behavior: by filtering the CSI data and selecting the subcarrier sensitive to the micro motion through the steps, a relatively regular waveform can be obtained. Fourier transforms can therefore be used to extract the tiny behavior frequencies, transforming the time domain signals into frequency domain signals.

Drawings

FIG. 1 is a pretreatment process involved in the process of the present invention

FIG. 2 shows a block diagram of a Hample filtering process before and after

FIG. 3 is a comparison graph before and after wavelet transformation

FIG. 4 shows the 30 subcarriers after the Hample filtering

FIG. 5 shows the variance values of 30 sub-carriers

FIG. 6 is an original image of 6 target sub-carriers

FIG. 7 shows a target subcarrier after noise processing

FIG. 8 is a time domain signal

FIG. 9 is a frequency domain signal

Detailed Description

The invention is further described with reference to the accompanying drawings in which:

the WiFi-based micro-behavior sensing technology is specifically divided into four steps:

(1) Micro-behavior modeling: the human respiration monitoring technology needs to extract a human respiration signal, and in order to accurately extract the human respiration signal and obtain the respiratory frequency, a human thoracic cavity needs to be modeled. Firstly, a Fresnel zone propagation theory is researched, a tiny behavior model is established on the basis of the Fresnel zone theory, and an experimental scheme is designed according to the model. When the micro-behavior is performed, a human body or an object is displaced, and a certain mapping relation exists between the displacement and the up-and-down fluctuation of the CSI signal. A human body or object is modeled as a sphere.

(2) Data preprocessing: as shown in fig. 1. Firstly, reading CSI data, extracting a CSI matrix, and then drawing original data of 30 subcarriers corresponding to each antenna. It can be seen that the raw data has noise interference in the acquisition environment, and outliers occur. The method comprises the steps that outlier removal is carried out by using a Hample filter, the Hample filter can filter amplitude values which are obviously different from other CSI measurement values, and noise suppression is carried out on collected CSI samples. After the outliers are removed, some sampling points are filtered out, and linear interpolation needs to be performed on the filtered signals to ensure the integrity of sampling. Sometimes, there is a sampling jitter phenomenon, which causes packet loss, and also requires interpolation processing. Linear interpolation is used to obtain data samples at evenly spaced time points, and the ordinate of the interpolated value is determined by a straight line determined by two points, as shown in fig. 2 for a comparison before and after the hash filtering process. After removing the abnormal value, the noise is filtered by adopting wavelet transform. The main noise source in the CSI raw data is caused by adaptive changes of transmission power and rate of the transmitting end and the receiving end or changes of the reference level. Conventional high frequency filters filter out many details that are useful for extracting the respiratory signal, and in order to preserve the edge information in the signal, a wavelet transform is used for noise filtering, as shown in fig. 3 for comparing before and after the wavelet transform.

(3) And (3) subcarrier selection: not every subcarrier of the 30 subcarriers per antenna may exhibit a regular breathing signal. Some subcarriers have small breathing amplitude and are not obvious in breathing signal, some subcarriers have large amplitude and present regular breathing signal, and some subcarriers have no periodic regularity, for example, fig. 4 shows 30 subcarriers after the hash filtering. Due to the difference in frequency, different subcarriers have different sensitivities to changes in the respiratory thorax. Different sub-carriers have different center frequencies, which causes different wavelengths and also causes different multipath effects and shadowing effects, so that different sub-carriers have different amplitudes. As can be seen from the CSI image, the larger the amplitude of the subcarrier, the more sensitive the subcarrier is to small movements. Therefore, the variance of each subcarrier can be calculated, and the variance is used to quantize the amplitude of the subcarrier. The sub-carrier with larger variance is selected as the target sub-carrier, and the threshold value is set to be 1.6 and is higher than the preset sub-carrier used for carrying out the respiratory frequency estimation. Through the screening of the set threshold value, the target subcarriers with the consistent period amplitude can be selected as

subcarriers

11, 12, 13, 26, 27 and 28. Fig. 5 shows the variance values of 30 subcarriers. Fig. 6 and 7 show the target subcarrier original image and the noise-filtered image selected by the threshold.

(4) Estimation of the frequency of the minor behavior: by filtering the CSI data and selecting the subcarrier sensitive to the micro-motion through the steps, a relatively regular respiration waveform can be obtained. Respiration is a periodic motion, so a fourier transform can be used to extract the respiration rate, converting the time domain signal into a frequency domain signal. The sampling time of the device is 120s, the sampling frequency is 100Hz, the total sampling number is 12000 data packets, and the frequency domain signal after Fourier transform can be obtained by setting parameters according to the time domain signal obtained by data preprocessing. The normal human breathing rate is 10-25bpm, corresponding to a frequency of 0.17-0.42Hz. Therefore, a bandpass filter may be used, with the bandpass cutoff frequency set to 0.17-0.42Hz, and a sliding window used to limit the frequency of small behaviors.

α＝argmax|FFT(x(1),...,x(n))|

Wherein, x (1),. And x (n) are selected target subcarriers. If the frequency of the micro-behavior of the experimental object is known to be beta, the estimation precision theta of the micro-behavior is as follows:

θ＝(1—|(α—β)/β|)×100％

the maximum frequency in the target subcarrier is 0.2353Hz, which translates to a breathing frequency of 14.118bpm. The micro-behavior rate of the experimental object is known to be 14bpm, and the estimated precision of the micro-behavior is calculated to be 97.24%. Fig. 8 is a time domain signal of the CSI signal after denoising, and a frequency domain signal of this breath obtained through fourier transform is shown in fig. 9.

Claims

1. A tiny behavior perception method based on WiFi signals is characterized by comprising the following four stages:

(1) Micro-behavior modeling: establishing a micro behavior model on the basis of Fresnel zone theory, wherein the establishment of the micro behavior model provides theoretical support for the acquisition of micro behavior data;

(2) Data preprocessing: firstly reading CSI data, extracting a CSI matrix, wherein original data obtained in the period can be inaccurate, and then sequentially carrying out outlier clearing, interpolation processing and noise transition to obtain regular micro-behavior signals;

(3) And (3) subcarrier selection: due to the difference of the frequency, different subcarriers have different sensitivities to small behavior changes, so that different subcarriers have differences in amplitude, the variance of each subcarrier is calculated, and the amplitude of the subcarriers is quantized by using the variances;

(4) Estimation of the minor behavior: through the steps, the CSI data are filtered, and the subcarrier sensitive to the micro motion is selected, so that a relatively regular micro-behavior waveform can be obtained, and the micro-behavior can be estimated through the waveform.

2. The method according to claim 1, wherein the method for sensing the micro-behaviors based on the WiFi signal comprises: the micro behavior modeling method comprises the steps that micro behavior signals need to be extracted, in order to accurately extract the micro behavior signals and obtain the respiratory frequency, micro behaviors of a human body or an object need to be modeled, a Fresnel zone propagation theory is firstly researched, a micro behavior model of the human body or the object is established on the basis of Fresnel, the design of a practical scheme zone theory is carried out according to the model, when the micro behaviors are carried out, the human body or the object can generate displacement along with behavior rhythm, and the human body or the object is modeled into a sphere.

3. The method of claim 1, wherein the method for sensing the micro-behavior based on the WiFi signal comprises: the method comprises the steps that data need to be processed in advance, firstly, CSI data are read, a CSI matrix is extracted, then original data of 30 subcarriers corresponding to each antenna are drawn, the fact that noise interference exists in an acquisition environment in the original data and an outlier appears can be seen, the outlier is removed by using Hample filtering, amplitude values which are obviously different from other CSI measurement values can be filtered out by a Hample filter, and noise suppression is conducted on an acquired CSI sample; after outliers are removed, some sampling points are filtered, linear interpolation needs to be carried out on the filtered signals to ensure the integrity of sampling, and sometimes, a sampling jitter phenomenon exists, which can cause data packet loss, and interpolation processing is also needed;

linear interpolation is utilized to obtain data samples of time points with uniform intervals, the ordinate of an interpolation value is determined by a straight line determined by two points, and after an abnormal value is removed, noise filtering is carried out by adopting sym8 wavelet basis and wavelet transformation of 5 layers of decomposition layers.

4. The method of claim 1, wherein the method for sensing the micro-behavior based on the WiFi signal comprises: due to the difference of frequencies, different sub-carriers have different sensitivities to the change of the respiratory thorax, the different central frequencies of the different sub-carriers cause different wavelengths and also cause different multipath benefits and shadow effects, so that the amplitudes of the different sub-carriers have difference; and calculating the variance of each subcarrier, quantizing the amplitude of the subcarrier by using the variance, selecting the subcarrier with larger variance as a target subcarrier, and setting the threshold value to be 1.6 which is higher than the preset subcarrier used for estimating the respiratory frequency.

5. The method of claim 1, wherein the method for sensing the micro-behavior based on the WiFi signal comprises: filtering CSI data and selecting a subcarrier sensitive to micro motion to obtain a regular micro behavior waveform; the method comprises the steps of extracting tiny behavior frequency by using Fourier transform, converting a time domain signal into a frequency domain signal, sampling time of equipment is 120s, the sampling frequency is 100Hz, the total number of sampling is 12000 data packets, setting parameters according to the time domain signal obtained by data preprocessing to obtain the frequency domain signal after the Fourier transform, and further achieving estimation of tiny behaviors.