Indoor positioning method based on two-dimensional sparse array signals

Abstract

The invention discloses an indoor positioning method based on two-dimensional sparse array signals. The method comprises the following steps: firstly, deploying a pair of equipment carrying a commercial Wi-Fi network card (Intel 5300) indoors, and configuring and acquiring channel state information (CSI) under multi-time snapshots in a form of'single transmitting antenna + three receiving antennas'; secondly, inhibiting phase distortion in the CSI data by using a linear fitting preprocessing method; thirdly, partitioning the total CSI data according to time snapshots, and inputting the total CSI data into a two-dimensional sparse iteration covariance estimation algorithm to estimate a multipath signal angle-time delay parameter; and finally, through a clustering algorithm, locking direct path features and completing accurate positioning. According to the method, the number of array elements of the receiving array is virtually expanded by using the subcarriers, the bottleneck of limited physical array elements of commercial Wi-Fi equipment is broken through, the positioning precision is greatly improved, the deployment cost is low, the compatibility is high, and the method is suitable for indoor positioning and other scenes.

Description

Technical Field

[0001] This invention pertains to indoor positioning technology, specifically relating to an indoor positioning method based on a two-dimensional sparse array signal. Background Technology

[0002] The advent of the 5G era has driven the rapid development of smart cities. People's demand for positioning and navigation is increasing daily, making location-based services (LBS) increasingly important. Currently, outdoor positioning technologies, such as the Global Positioning System (GPS) and cellular networks, are mature enough to meet users' daily needs for sub-meter level positioning, route navigation, and logistics tracking. However, due to numerous obstacles and obstructions in indoor environments, satellite or cellular network signals are significantly attenuated, rendering mature outdoor positioning systems ineffective indoors. In today's world, with the continuous advancement of modernization and intelligence, most people's lives and work are conducted indoors, making the demand for indoor location services even more urgent. In recent years, due to the increasing demand for indoor positioning across various industries, indoor positioning technologies based on Wi-Fi, infrared, ultrasonic, Bluetooth, and Zigbee signals have developed significantly. Among them, Wi-Fi-based indoor positioning technology has gained widespread attention due to its low deployment cost and high positioning accuracy, and has gradually become the mainstream indoor positioning technology.

[0003] Wi-Fi-based positioning algorithms can be broadly categorized into two types: one is based on traditional signal processing techniques to extract relevant positioning features for location estimation; the other utilizes artificial intelligence (AI) technology for data processing and location estimation. Compared to AI-based location estimation methods, traditional signal processing techniques for feature parameter estimation offer several advantages: First, they do not require extensive training with large amounts of data and can be deployed on existing commercial Wi-Fi infrastructure without additional equipment modifications. Second, parameter estimation methods are not dependent on fixed features specific to a particular scenario and exhibit good adaptability to dynamically changing factors in real-world environments. Finally, while the integration with AI technology leads to a loss of interpretability in the overall positioning system, the steps of parameter estimation strictly adhere to physical formulas and mathematical logic, allowing for traceability and explanation of the estimation process and error sources, facilitating troubleshooting and optimization. Despite these advantages, indoor scenarios often suffer from severe path loss and complex multipath interference. Commercial Wi-Fi devices typically have insufficient array elements to meet the identification needs of real-world indoor multipath scenarios, resulting in difficulties in correctly identifying the LOS path.

[0004] To address the aforementioned problems, this invention proposes an indoor positioning method based on two-dimensional sparse array signals. This method virtually expands the number of array elements by combining physical antennas and subcarriers, achieving a number of elements greater than the number of multipath paths in the indoor environment. Furthermore, a super-resolution algorithm, two-dimensional sparse iterative covariance estimation, is used to estimate the angle and time delay of the LOS path signal, ultimately achieving precise positioning. Summary of the Invention

[0005] The purpose of this invention is to provide an indoor positioning method based on two-dimensional sparse array signals. This method utilizes actual antennas and subcarriers to virtually extend the array, and employs a super-resolution algorithm to estimate the path angle and time delay parameters, achieving precise positioning. This method does not require a large dataset for training, making it very lightweight. Furthermore, this solution only requires simple commercial Wi-Fi equipment, facilitating deployment and implementation. This method overcomes the hardware limitations of equipment, achieving high resolution while effectively controlling complexity and cost, and ensuring sub-meter-level positioning accuracy.

[0006] The indoor positioning method based on two-dimensional sparse array signals described in this invention mainly includes the following steps:

[0007] Step 1: Prepare a pair of devices equipped with commercial Wi-Fi network cards (Intel 5300). One device will be used as a signal transmitter, equipped with one transmitting antenna, and the other will be used as a receiver, with three receiving antennas arranged in a uniform linear array. The devices will be equipped with the Linux 802.11n CSI Tool to collect CSI measurements. The devices will be set to a 5.32GHz-64 channel operating mode, with a data packet transmission rate of 1000Hz.

[0008] Step 2: The transmitting device broadcasts data packets. After reflection from indoor walls and other obstacles, multiple paths with different angles, including the LOS path, are formed and mapped onto a ULA receiving array with M elements. The entire angle space is divided into K grid points, and the k-th angle grid point (k∈[K]) θ k The phase offset on the m-th antenna (m∈[M]) is -2×π×d×(m-1)×sin(θ) k )×f / c, where d is the antenna spacing, f is the signal frequency, and c is the speed of light. Based on the relationship between the ULA arrays, the phase shift of a single path angle can be expressed as: in The phase shift caused by the time delay is mainly reflected in different subcarriers of the same antenna. The possible time delay range is divided into S grid points, and the time delay τ corresponding to N subcarriers is...s The phase shift is in f δ This refers to the frequency spacing of the subcarriers. To expand the number of pair elements, the dimensions need to be combined to form a two-dimensional phase offset that includes angle and time delay. At this point, the number of array elements is expanded from M to M0 = M × N, which can provide a higher resolution for the model.

[0009] Step 3: Construct a two-dimensional sparse iterative covariance estimation algorithm model based on an extended two-dimensional virtual array. This includes the following steps:

[0010] Step 3 (a): Based on the virtual extended array and two-dimensional grid phase offset form obtained in Step 2 Modeling is performed on the received signal y(t) captured in a single time snapshot. The two-dimensional grid points correspond to a K×S two-dimensional overcomplete time-delay dictionary A. The elements correspond to two-dimensional phase shifts. The sparse signal matched with A is x(t) = [x1(t), x2(t), ..., x...]. K×S (t)] Τ x i (t) has a non-zero value only at the (θ,τ) grid point where the real signal exists; the x values ​​at other locations are non-zero. i The values ​​of y(t) are all approximately 0. Thus, the signal received by the receiver at time t can be expressed as y(t) = Ax(t) + n(t), where n(t) is noise (t∈T, T is the number of time snapshots).

[0011] Step 3 (II): Construct the computational model of the two-dimensional sparse iterative covariance estimation algorithm. In Step 3 (I), the signal model y(t) was obtained, and the covariance matrix can be expressed as... The above formula can be equivalent to: After considering σ as another form of p, that is I is an identity matrix of dimension M0. To estimate {p} i The SPICE algorithm is needed for sparse reconstruction, i.e., solving the problem. Where the parameter ω i The expression is R xx It is the sample covariance matrix of the signal. T is the number of time snapshots. By solving the above formula, the path position where the actual signal exists can be obtained, and the relevant angle and time delay information can be obtained by looking up the information in the overcomplete dictionary A.

[0012] Step 4: The directly acquired CSI signal contains various phase noises caused by asynchrony between transceivers and hardware defects, such as timing offset (TO) and carrier frequency offset (CFO). The presence of these noises makes it impossible to directly estimate signal parameters from the raw CSI signal. Since phase noise is linearly distributed across different subcarriers, a linear fitting method can be used to remove it. The phase offset on the nth subcarrier is expressed as -2πf. δ (n-1)τ phy The CSI data, after being corrected by linear fitting, can then be input into the model in step three to estimate the channel parameters.

[0013] Step 5: Divide the data into blocks according to time, make multiple estimates, and cluster the final results to find the optimal solution. This includes the following steps:

[0014] Step 5 (a): Divide the overall time snapshot T into multiple sub-snapshots. Choose an appropriate window size and divide T into n sub-snapshots of size T. c The sub-snapshot (T=[T) c1 ,T c2 ,...,T cn Meanwhile, the received signal model obtained in step three (a) can be written as Y = [Y1, Y2, ..., Y]. n ], where Y i =[y1(t),y2(t),...,y c (t)].

[0015] Step 5 (II): Use the two-dimensional sparse iterative covariance estimation algorithm to calculate the data for each sub-snapshot, and obtain multiple {p} i The estimation of Y is performed using the two-dimensional sparse iterative covariance estimation algorithm model obtained in step three (two). i Substituting these values ​​into the calculation, we obtain R and R in each sub-signal. xx And calculate {p} according to the formula. i This yields the set of angle-time delay information {(θ,τ)} corresponding to the actual path (including direct and reflected paths). i}

[0016] Step 5 (III): Use clustering algorithms to determine the direct path in space. For the results of cross-time snapshots, the trend of change of the direct path is smaller than that of the reflection path. Therefore, by clustering the results of multiple sub-data sets, the trend of change of different paths can be obtained, thereby finding the direct path. Using the Gaussian mean clustering algorithm, the n estimated results are clustered to find the angle-delay information (θ) corresponding to the direct path. los,τ los ).

[0017] Step 6: Based on the obtained direct access to the corresponding parameter information (θ) los ,τ los To achieve precise positioning.

[0018] Beneficial effects

[0019] This invention proposes an indoor positioning method based on two-dimensional sparse array signals, starting from the structural degree of freedom extension of antenna arrays. First, leveraging the OFDM technology underlying Wi-Fi signals, subcarriers are used to expand the actual array size, improving the device's signal resolution capability. Second, a linear fitting algorithm is used to remove phase offsets from the original CSI signal, obtaining usable CSI data. Third, a high-resolution two-dimensional sparse iterative covariance estimation algorithm is proposed, using sub-data information from multiple time snapshots to perform sparse reconstruction estimation of the covariance matrix R, obtaining an estimated solution for the angle-delay information (θ, τ) of the sparse path. Finally, a Gaussian mean clustering algorithm is used to cluster the estimated solutions obtained from multiple sub-data, and based on the principle that the change across time snapshots of the direct path is small, the angle-delay parameter information of the signal direct path is obtained, thereby achieving accurate positioning. Attached Figure Description

[0020] Figure 1 This is a flowchart of the present invention;

[0021] Figure 2 Diagram of the ULA array receiving signal model; Detailed Implementation Plan

[0022] The present invention will be further described below with reference to the accompanying drawings:

[0023] like Figure 1 The indoor positioning method based on two-dimensional sparse array signals shown includes the following steps:

[0024] Step 1: Prepare a pair of devices equipped with commercial Wi-Fi network cards (Intel 5300). One device will be used as a signal transmitter, equipped with one transmitting antenna, and the other will be used as a receiver, with three receiving antennas arranged in a uniform linear array. The devices will be equipped with the Linux 802.11n CSI Tool to collect CSI measurements. The devices will be set to a 5.32GHz-64 channel operating mode, with a data packet transmission rate of 1000Hz.

[0025] Step 2: The transmitting device broadcasts data packets. After reflection from indoor walls and other obstacles, multiple paths with different angles, including the LOS path, are formed and mapped onto a ULA receiving array with M elements. The entire angle space is divided into K grid points, and the k-th angle grid point (k∈[K]) θ k The phase offset on the m-th antenna (m∈[M]) is -2×π×d×(m-1)×sin(θ) k )×f / c, where d is the antenna spacing, f is the signal frequency, and c is the speed of light. Based on the relationship between the ULA arrays, the phase shift of a single path angle can be expressed as: in The phase shift caused by the time delay is mainly reflected in different subcarriers of the same antenna. The possible time delay range is divided into S grid points, and the time delay τ corresponding to N subcarriers is... s The phase shift is in f δ This refers to the frequency spacing of the subcarriers. To expand the number of pair elements, the dimensions need to be combined to form a two-dimensional phase offset that includes angle and time delay. At this point, the number of array elements is expanded from M to M0 = M × N, which can provide a higher resolution for the model.

[0026] Step 3: Construct a two-dimensional sparse iterative covariance estimation algorithm model based on an extended two-dimensional virtual array. This includes the following steps:

[0027] Step 3 (a): Based on the virtual extended array and two-dimensional grid phase offset form obtained in Step 2 Modeling is performed on the received signal y(t) captured in a single time snapshot. The two-dimensional grid points correspond to a K×S two-dimensional overcomplete time-delay dictionary A. The elements correspond to two-dimensional phase shifts. The sparse signal matched with A is x(t) = [x1(t), x2(t), ..., x...]. K×S (t)] Τ x i (t) has a non-zero value only at the (θ,τ) grid point where the real signal exists; the x values ​​at other locations are non-zero. i The values ​​of y(t) are all approximately 0. Thus, the signal received by the receiver at time t can be expressed as y(t) = Ax(t) + n(t), where n(t) is noise (t∈T, T is the number of time snapshots).

[0028] Step 3 (II): Construct the computational model of the two-dimensional sparse iterative covariance estimation algorithm. In Step 3 (I), the signal model y(t) was obtained, and the covariance matrix can be expressed as... The above formula can be equivalent to: After considering σ as another form of p, that is I is an identity matrix of dimension M0. To estimate {p} i The SPICE algorithm is needed for sparse reconstruction, i.e., solving the problem. Where the parameter ω i The expression is R xx It is the sample covariance matrix of the signal. T is the number of time snapshots. By solving the above formula, the path position where the actual signal exists can be obtained, and the relevant angle and time delay information can be obtained by looking up the information in the overcomplete dictionary A.

[0029] Step 4: The directly acquired CSI signal contains various phase noises caused by asynchrony between transceivers and hardware defects, such as timing offset (TO) and carrier frequency offset (CFO). The presence of these noises makes it impossible to directly estimate signal parameters from the raw CSI signal. Since phase noise is linearly distributed across different subcarriers, a linear fitting method can be used to remove it. The phase offset on the nth subcarrier is expressed as -2πf. δ (n-1)τ phy The CSI data, after being corrected by linear fitting, can then be input into the model in step three to estimate the channel parameters.

[0030] Step 5: Divide the data into blocks according to time, make multiple estimates, and cluster the final results to find the optimal solution. This includes the following steps:

[0031] Step 5 (a): Divide the overall time snapshot T into multiple sub-snapshots. Choose an appropriate window size and divide T into n sub-snapshots of size T. c The sub-snapshot (T=[T) c1 ,T c2 ,...,T cn Meanwhile, the received signal model obtained in step three (a) can be written as Y = [Y1, Y2, ..., Y]. n ], where Y i =[y1(t),y2(t),...,y c (t)].

[0032] Step 5 (II): Use the two-dimensional sparse iterative covariance estimation algorithm to calculate the data for each sub-snapshot, and obtain multiple {p} i The estimation of Y is performed using the two-dimensional sparse iterative covariance estimation algorithm model obtained in step three (two). i Substituting these values ​​into the calculation, we obtain R and R in each sub-signal.xx And calculate {p} according to the formula. i This yields the set of angle-time delay information {(θ,τ)} corresponding to the actual path (including direct and reflected paths). i}

[0033] Step 5 (III): Use clustering algorithms to determine the direct path in space. For the results of cross-time snapshots, the trend of change of the direct path is smaller than that of the reflection path. Therefore, by clustering the results of multiple sub-data sets, the trend of change of different paths can be obtained, thereby finding the direct path. Using the Gaussian mean clustering algorithm, the n estimated results are clustered to find the angle-delay information (θ) corresponding to the direct path. los ,τ los ).

[0034] Step 6: Based on the obtained direct access to the corresponding parameter information (θ) los ,τ los To achieve precise positioning.

Claims

1. An indoor positioning method based on two-dimensional sparse array signals, characterized by the following steps: Step 1: Prepare a pair of devices equipped with commercial Wi-Fi network cards (Intel 5300). One device will be used as a signal transmitter, equipped with one transmitting antenna, and the other will be used as a receiver, with three receiving antennas arranged in a uniform linear array. The devices will be equipped with Linux 802.11n CSITool to collect CSI measurements. The devices will be set to a 5.32GHz-64 channel operating mode, with a data packet transmission rate of 1000Hz. Step 2: The transmitting device broadcasts data packets. After reflection from indoor walls and other obstacles, multiple paths with different angles, including the LOS path, are formed and mapped onto a ULA receiving array with M elements. The entire angle space is divided into K grid points, and the k-th angle grid point (k∈[K]) θ k The phase offset on the m-th antenna (m∈[M]) is -2×π×d×(m-1)×sin(θ) k )×f / c, where d is the antenna spacing, f is the signal frequency, and c is the speed of light. Based on the relationship between the ULA arrays, the phase shift of a single path angle can be expressed as: in The phase shift caused by the time delay is mainly reflected in different subcarriers of the same antenna. The possible time delay range is divided into S grid points, and the time delay τ corresponding to N subcarriers is... s The phase shift is in f δ This refers to the frequency spacing of the subcarriers. To expand the number of pair elements, the dimensions need to be combined to form a two-dimensional phase offset that includes angle and time delay. At this point, the number of array elements is expanded from M to M0 = M × N, which can provide a higher resolution for the model. Step 3: Build a two-dimensional sparse iterative covariance estimation algorithm model based on the extended two-dimensional virtual array. Step 4: The directly acquired CSI signal contains various phase noises caused by asynchrony between transceivers and hardware defects, such as timing offset (TO) and carrier frequency offset (CFO). The presence of these noises makes it impossible to directly estimate signal parameters from the raw CSI signal. Since phase noise is linearly distributed across different subcarriers, a linear fitting method can be used to remove it. The phase offset on the nth subcarrier is expressed as -2πf. δ (n-1)τ phy The CSI data, after being corrected by linear fitting, can then be input into the model in step three to estimate the channel parameters. Step 5: Divide the data into blocks according to time, make multiple estimates, and cluster the final results to find the optimal solution. Step 6: Based on the obtained parameter information (θ) corresponding to the direct path los ,τ los To achieve precise positioning.

2. The indoor positioning method based on two-dimensional sparse array signals according to claim 1, characterized in that... Step three includes the following steps: Step 3: Construct a two-dimensional sparse iterative covariance estimation algorithm model based on an extended two-dimensional virtual array. This includes the following steps: Step 3 (a): Based on the virtual extended array and two-dimensional grid phase offset form obtained in Step 2 Modeling is performed on the received signal y(t) captured in a single time snapshot. The two-dimensional grid points correspond to a K×S two-dimensional overcomplete time-delay dictionary A. The elements correspond to two-dimensional phase shifts. The sparse signal matched with A is x(t) = [x1(t), x2(t), ..., x...]. K×S (t)] Τ x i (t) has a non-zero value only at the (θ,τ) grid point where the real signal exists; the x values ​​at other locations are non-zero. i The values ​​of y(t) are all approximately 0. Thus, the signal received by the receiver at time t can be expressed as y(t) = Ax(t) + n(t), where n(t) is noise (t∈T, T is the number of time snapshots). Step 3 (II): Construct the computational model of the two-dimensional sparse iterative covariance estimation algorithm. In Step 3 (I), the signal model y(t) was obtained, and the covariance matrix can be expressed as... The above formula can be equivalent to: After considering σ as another form of p, that is I is an identity matrix of dimension M0. To estimate {p} i The SPICE algorithm is needed for sparse reconstruction, i.e., solving the problem. Where the parameter ω i The expression is R xx It is the sample covariance matrix of the signal. T is the number of time snapshots. By solving the above formula, the path position where the actual signal exists can be obtained, and the relevant angle and time delay information can be obtained by looking up the information in the overcomplete dictionary A.

3. The indoor positioning method based on two-dimensional sparse array signals according to claim 1, characterized in that... Step five includes the following steps: Step 5: Divide the data into blocks according to time, make multiple estimates, and cluster the final results to find the optimal solution. This includes the following steps: Step 5 (a): Divide the overall time snapshot T into multiple sub-snapshots. Choose an appropriate window size and divide T into n sub-snapshots of size T. c The sub-snapshot (T=[T) c1 ,T c2 ,...,T cn Meanwhile, the received signal model obtained in step three (a) can be written as Y = [Y1, Y2, ..., Y]. n ], where Y i =[y1(t),y2(t),...,y c (t)]. Step 5 (II): Use the two-dimensional sparse iterative covariance estimation algorithm to calculate the data for each sub-snapshot, and obtain multiple {p} i The estimation of Y is performed using the two-dimensional sparse iterative covariance estimation algorithm model obtained in step three (two). i Substituting these values ​​into the calculation, we obtain R and R in each sub-signal. xx And calculate {p} according to the formula. i This yields the set of angle-time delay information {(θ,τ)} corresponding to the actual path (including direct and reflected paths). i } Step 5 (III): Use clustering algorithms to determine the direct path in space. For the results of cross-time snapshots, the trend of change of the direct path is smaller than that of the reflection path. Therefore, by clustering the results of multiple sub-data sets, the trend of change of different paths can be obtained, thereby finding the direct path. Using the Gaussian mean clustering algorithm, the n estimated results are clustered to find the angle-delay information (θ) corresponding to the direct path. los ,τ los ).

Images