Backscattering label deployment method based on channel knowledge graph

Abstract

The invention discloses a backscattering label deployment method based on a channel knowledge graph. The backscattering label deployment method comprises the following steps: acquiring actual channel data in an offline measurement environment; constructing a spatial perception model of wireless signal propagation by adopting a spatial loss field modeling method; modeling shadow fading by using a Gaussian process regression method, deducing shadow loss distribution at any position, and finally constructing a wireless channel knowledge graph; analyzing interference distribution according to the atlas, selecting an area with the minimum interference as a deployment position of a backscattering label, and selecting a proper precoding strategy according to link characteristics; in the actual deployment stage, the communication quality of each position is evaluated through the signal to interference plus noise ratio, the deployment position and system parameters are adjusted, and accurate deployment of the backscattering label is completed. According to the method, backscattering label deployment is optimized through the wireless channel knowledge graph, accurate interference prediction and environment adaptability are provided, and an efficient and low-cost solution is provided for the future 6G passive Internet of Things.

Description

Technical Field

[0001] The present invention belongs to the field of wireless communication technology, and in particular relates to a backscatter tag deployment method based on a channel knowledge graph. Background Art

[0002] In the 6G era, the Internet of Things is considered one of the most important application scenarios. 6G requires breakthroughs in high-density device connections, large data transmission, and low latency. The Internet of Things often has requirements such as low power consumption, low cost, and intermittent data output. In order to reduce the power consumption and cost of IoT devices, backscatter tags have been introduced as a core technology. Backscatter tags achieve low-power communication by modulating information onto an external transmitted signal (such as a continuous wave signal from a base station) and decoding the signal at the receiving end. However, because backscatter tags rely on external transmitted signals for information reflection, their signals are susceptible to interference sources in the surrounding environment, especially in dense wireless communication environments. For example, interference from other devices, base stations, or backscatter tags may significantly affect their communication performance and reduce the overall efficiency of the system.

[0003] To minimize this interference, backscatter tag deployment strategies must be carefully designed. In dense wireless communication environments, interference can arise not only from adjacent base stations but also from other backscatter tags, wireless devices, and even multipath propagation. Because backscatter tags rely on external signals for information transmission, these interference sources can cause significant attenuation or distortion of the received signal, impacting communication quality. Therefore, precisely selecting backscatter tag deployment locations to avoid areas of high interference is crucial for improving system performance. Summary of the Invention

[0004] Purpose of the Invention: This invention aims to provide a backscatter tag deployment method based on a channel knowledge graph. By optimizing backscatter tag deployment through a wireless channel knowledge graph, it provides accurate interference prediction and environmental adaptability, and offers an efficient, low-cost, and low-power solution for future 6G passive IoT.

[0005] Technical solution: A backscatter tag deployment method based on a channel knowledge graph of the present invention includes the following steps:

[0006] Step 101: Construct a MIMO communication experimental scenario including backscatter tags, and in a controlled offline measurement environment, use a signal generator to simulate base station signal transmission and a receiver to receive signals to collect actual wireless channel data for subsequent modeling and analysis.

[0007] Step 102: Using the spatial loss field modeling method, the collected actual wireless channel data is used to construct a spatial perception model that describes the propagation characteristics of wireless signals.

[0008] Step 103: Using the Gaussian process regression method, the shadow fading of the spatial position is continuously modeled by introducing a residual term, and the shadow loss distribution of each position is inferred, thereby constructing a wireless channel knowledge graph;

[0009] Step 104: Based on the constructed wireless channel knowledge graph, taking into account the path loss model and environmental characteristics, an interference distribution map of the wireless environment is obtained. Based on the interference distribution map of the target area, the area with the least interference in the environment is analyzed, and areas with low interference power and high channel stability are selected as candidate deployment locations for backscatter tags.

[0010] Step 105: Evaluate the channel quality of each candidate deployment location using the signal-to-interference-and-noise ratio (SINR) as a performance evaluation metric. Addressing the issue of low SINR values ​​for some backscatter tag links, select a precoding strategy that matches the backscatter tag link to optimize the candidate deployment location and precoding strategy.

[0011] Step 106: Based on the optimized candidate deployment locations and precoding strategies, during the actual system deployment phase, the signal-to-interference-and-noise ratios of various locations in the target area are compared to evaluate the communication quality of the backscatter tag link. The deployment locations are further adjusted to complete the precise deployment of the backscatter tags.

[0012] Furthermore, step 101 is specifically as follows: the MIMO communication experimental scenario includes a base station, a plurality of backscatter tags and a plurality of users; wherein, the base station is denoted as BS, located at a fixed position (0,0) in a two-dimensional coordinate system, and the backscatter tag set is denoted as Actual channel data is obtained in an offline measurement environment. Backscatter tags and user terminals are deployed in the set target area. A signal generator is used to generate a detection signal to simulate the transmission of the base station. The backscatter tag is in a reflective working state. The user equipment receiving module synchronously collects the received signal characteristics and location information. The user terminal moves to each measurement point according to the predetermined trajectory and records the received channel parameters including signal power, path loss, shadow fading, multipath component, arrival angle, and transmission angle. Each set of measurement data is stored synchronously with the corresponding spatial position to form a standardized measurement data pair. After repeated measurements, the target deployment area is covered. After the acquisition is completed, the data set is detected and cleaned for anomalies to remove invalid or abnormal measurement points, and finally an offline measurement data set is obtained for subsequent construction of the wireless channel knowledge graph.

[0013] Furthermore, step 102 is specifically as follows: dividing the target area into N x ×N y The size of each grid cell is Δx and Δy, and the discretized grid point set of the target area is:

[0014] G={(iΔx,jΔy)∣i=1,...,N x ,j=1,...,N y}

[0015] Where Δx and Δy are the sizes of the grid cells in the x and y directions, which control the accuracy of the grid resolution. i and j are the indices of the grid cells in the x and y directions.

[0016] The spatial loss field model is used to model the environment. The spatial loss field model of the wireless channel uses a normalized ellipse model to describe the path loss distribution between the transmitter and the receiver. The normalized ellipse model assumes that all positions within the ellipsoid formed by the two points of transmission and reception have an impact on the shadow fading of the link; when a point is within the ellipse with the transmitter and receiver as the focus, it is considered that the point contributes to the signal propagation, otherwise the impact is negligible; for all points within the ellipse, when a position is within the ellipse, its path weight is constant Where φ1 is the direct path length, and w = 0 outside the area. Considering the influence of free space propagation and environmental shadows, a path loss model is established, and the logarithmic distance loss model is used to represent the free space loss: the reference path loss β is defined at the reference distance d0, and the influence of the propagation distance is represented by the path loss exponent α, α>0, d represents the distance from the transmitter to the receiver, and the free space path loss is expressed as:

[0017]

[0018] The shadow fading term is introduced to reflect the additional loss caused by obstacles. Based on the aforementioned discretized spatial loss field, the shadow loss is modeled as a weighted superposition of the loss contributions of each grid:

[0019]

[0020] Among them, x i and x j Represent the locations of the transmitting and receiving points respectively, φ1(x i ,x j )=|x i -x j | is the direct path distance, φ2(x p ;x i ,x j )=|x p -x i |+|x p -x j | indicates passing through the middle point x p The diffraction path length, weight function w(φ1,φ2) is based on the normalized ellipse model and is defined as: if φ2≤φ1+Δ, Δ is half wavelength, then w is Otherwise w=0; δ p is the loss value of the p-th spatial unit, which is the attenuation caused by obstacles; the shadow fading term S(x i ,x j ) is equivalent to the loss value δ of all grid cells passed by the signal propagation path within the first Fresnel ellipse area. p Sum; combining free space loss and shadow loss, we get the total path loss model of the link:

[0021]

[0022] Where β is the reference path loss at the reference distance d0, α is the path loss exponent, and δ p is the loss value of the pth spatial unit, and w(φ1, φ2) is the weight factor when the signal path passes through the grid unit.

[0023] Furthermore, step 103 is specifically as follows: the residual reflects random shadow fading, and is modeled and predicted using a Gaussian process, thereby introducing data-driven correction into the model; after extracting the residual and analyzing its distribution characteristics, the residual term is assumed to approximately obey a zero-mean Gaussian random field, whose covariance is determined by the distance relationship between spatial positions, and a Gaussian process regression model is introduced to fit the residual, and a channel map is constructed for the coverage area of ​​a fixed base station. The position (x, y) of the receiving terminal is selected as input, and the output is the loss residual value r(x) of the corresponding position; after mean normalization, the Gaussian process establishes a mapping model f:(x, y)→r from the position coordinate to the residual value; an isotropic Gaussian kernel function is selected and the correlation between the two positions is calculated using the Euclidean distance between the two positions, which is in the form of:

[0024]

[0025] Among them, l is called the length scale, which determines the decay rate of the correlation between spatial positions. is the prior variance of the channel residual; when the distance |x p -x q |Close to 0, the kernel function value is close to Indicates that the residuals are highly correlated; as the distance increases, the kernel function exponential term approaches zero, the correlation decreases, and the residuals of distant points are considered to be almost uncorrelated; by adjusting l to match the relevant distance scale of shadow fading in the actual environment, in Gaussian process regression, assuming that the residuals obey the zero-mean Gaussian process prior, the posterior prediction is also Gaussian distributed, and the predicted mean of Gaussian process regression is expressed as:

[0026]

[0027] Where r=[r(x1),...,r(x N )]T is the training residual vector, K is the N×N training sample covariance matrix, and its element K ij =k(x i ,x j ) represents the sample x i with x j The correlation between x * represents any new test position in the region, k * =[k * (x * ,x1),...,k * (x * ,x N )] T Represents the new position x * and the covariance vector between all training points, The noise variance is the residual prediction for the new position, which is the weighted average of the residual values ​​of each observation in the training set. The weight is determined by the correlation between the new point and each known point calculated by the kernel function, and is normalized by the inverse matrix of the autocorrelation of the training points. The residual correction term obtained by the Gaussian process prediction is combined with the previous path loss benchmark model to give an estimated value of the path loss for any position in the continuous space. For any point x in the region, * , and its path loss is estimated as:

[0028]

[0029] Among them, L total (x * ) represents the total path loss model of the link, is the predicted mean of the Gaussian process regression.

[0030] Furthermore, step 104 is specifically as follows: based on the constructed wireless channel knowledge graph, considering the path loss model and environmental characteristics, obtaining the interference distribution of the wireless environment, and the path loss calculation formula is as follows:

[0031] P signal (x r )=P transmit -PL(x r )

[0032] Among them, P signal (x r ) is the received signal power, P transmit is the transmission power, PL(x r ) is the distance from the transmitting point to the receiving point x r Path loss;

[0033] Then calculate the interference signal power:

[0034] P interf (x r ,x i )=P transmit (x i )-PL(x r ,x i )

[0035] Among them, P interf (x r ,x i ) is the interference source x i For receiving point x r The interference power, P transmit (x i ) is the interference source x i The transmission power, PL(x r ,x i ) is the interference source x i To receiving point x r Path loss;

[0036] Based on the interference distribution and the strength of the target signal, the SINR at each location is calculated to assess the interference level at each location. The signal-to-interference-and-noise ratio is calculated as follows:

[0037]

[0038] Among them, P signal (x r ) is the received power of the target signal, is the total interference source to the receiving point x r The sum of the interference powers, P noise is the noise power;

[0039] For each position x r , calculate its signal to interference noise ratio. Get the interference distribution map of the target area.

[0040] Furthermore, step 105 specifically includes: evaluating the channel quality of each deployment point using the signal-to-interference-and-noise ratio as a performance evaluation indicator. To address the problem of low signal-to-interference-and-noise ratio values ​​for links to some backscatter tags, a hybrid MRC precoding matrix is ​​selected. The hybrid MRC precoding matrix is ​​expressed as:

[0041] v k =h k +g k

[0042] Among them, h k represents the direct link channel matrix, g k The backscatter tag link channel matrix is ​​represented by first decoding the direct link signal and then using the successive interference cancellation technique. The formula is as follows:

[0043]

[0044] Among them, K represents the number of users, y k Indicates the received signal, p i Indicates the transmit power, v k Represents the receiving combination vector, h i represents the direct link channel matrix, represents the direct link interference signal, It represents its average value, that is, the removed direct link interference signal, s i Indicates direct link information symbol, c i represents the backscatter tag link information symbol, Γ represents the reflection coefficient of the backscatter tag, N k Represents a noise signal; it removes some interference from the direct link when decoding the backscatter tag signal, improving the signal-to-interference-noise ratio of the backscatter tag link.

[0045] Furthermore, in step 106, the signal-to-interference-noise ratios (SINRs) of various locations in the target area are compared, and backscatter tags are preferentially deployed at locations with higher SINRs.

[0046] The present invention further discloses a computer device, comprising a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method of the present invention.

[0047] The present invention further discloses a computer-readable storage medium having a computer program / instruction stored thereon, which implements the steps of the method of the present invention when the computer program / instruction is executed by a processor.

[0048] The present invention further discloses a computer program product, comprising a computer program / instruction, which implements the steps of the method of the present invention when executed by a processor.

[0049] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages:

[0050] 1. The present invention accurately predicts and analyzes the interference distribution in the target area through the wireless channel knowledge graph, thereby selecting the deployment location with the least interference, ensuring the stability and reliability of the communication link.

[0051] 2. The present invention can provide environmental perception capabilities for complex environments, enabling backscatter tags to dynamically adapt to different environmental conditions (such as multipath effects, obstructions, etc.), effectively optimizing signal propagation and reception effects.

[0052] 3. The technical solution of the present invention provides an efficient, low-cost, and low-power solution for large-scale passive IoT device connections in future 6G networks.

[0053] 4. By performing the above estimations for all locations across the entire area, the present invention constructs a continuous spatial wireless channel map. Using location coordinates as indexes, it maps the large-scale channel loss values ​​corresponding to each location. The channel knowledge graph can be considered a geographic database that stores the path loss at each location. This knowledge graph is continuous and context-aware, offering higher accuracy in complex environments than relying solely on empirical formulas or discrete measurement points. Physical-based loss predictions provide a fundamental estimate of distance attenuation and major obstacle occlusion. Gaussian process data-driven correction of model errors allows the estimated map to approximate measured values. By integrating a spatial loss field model with Gaussian process regression, the present invention not only comprehensively describes the large-scale loss characteristics of wireless channels but also effectively infers the channel state at unknown locations based on limited measurement data, thereby constructing a precise and continuous wireless channel knowledge graph.

[0054] 5. Hybrid MRC precoding of the present invention: This method first decodes the direct link signal and then uses successive interference cancellation to remove the interference signal of the direct link. This improves the signal-to-interference-noise ratio when decoding the backscatter tag signal, thereby significantly improving the signal quality of the backscatter tag link. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 This is a flow chart of a backscatter tag deployment method based on a channel knowledge graph according to an embodiment of the present invention;

[0056] Figure 2 This is a framework diagram of a backscatter tag deployment method based on a channel knowledge graph in an embodiment of the present invention. DETAILED DESCRIPTION

[0057] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0058] 1. The present invention provides a backscatter tag deployment method based on channel knowledge graph, such as Figure 1 As shown, the method includes the following steps:

[0059] Step 101: Construct a MIMO communication experimental scenario including backscatter tags, and in a controlled offline measurement environment, use a signal generator to simulate base station signal transmission and a receiver to receive signals to collect actual wireless channel data for subsequent modeling and analysis.

[0060] Step 102: To construct an environment-aware wireless channel knowledge graph, first, a spatial loss field modeling method is used to construct a spatial perception model that describes wireless signal propagation characteristics based on the collected limited offline measurement data.

[0061] Step 103: Further utilize the Gaussian process regression method to reflect the impact of random shadow fading on model prediction by introducing a residual term, continuously model the shadow fading of spatial locations, and infer the shadow loss distribution at each location, thereby constructing a wireless channel knowledge graph;

[0062] Step 104: Based on the constructed wireless channel knowledge graph, the path loss model and environmental characteristics (such as obstacles and multipath effects) are considered to obtain the interference distribution of the wireless environment. Based on the interference distribution of the target area, the area with the least interference in the environment is analyzed, and areas with low interference power and high channel stability are preferentially selected as candidate deployment locations for backscatter tags.

[0063] Step 105: Evaluate the channel quality of each deployment point using the signal-to-interference-and-noise ratio (SINR) as a performance evaluation metric. Addressing the issue of low SINR values ​​for some backscatter tag links, select a precoding strategy that matches the backscatter tag links to improve the reliability and coverage of the backscatter communication links.

[0064] Step 106: Based on the optimized deployment locations and precoding strategy, during the actual system deployment phase, the signal-to-interference-and-noise ratios (SINRs) at various locations in the target area are compared to evaluate the communication quality of the backscatter tag link. The deployment locations are further adjusted to achieve precise deployment of the backscatter tags, thereby ensuring the stability and efficiency of the communication link.

[0065] 2. Backscatter Tag Deployment Architecture Based on Channel Knowledge Graph

[0066] The backscatter tag deployment architecture based on the channel knowledge graph of this embodiment is as follows Figure 2 As shown, the illustrated scenario includes one base station (BS), L backscatter tags, and K user devices. The entire system consists of modules 201, 202, 203, 204, 205, and 206. Module 201 is the base station (BS), responsible for sending data; module 202 is the user device (UE), which receives data; module 203 is a dedicated signal generator, used to simulate base station transmission signals in a dedicated offline environment; module 204 is the radiated signal; and module 205 is the backscatter tag, which modulates and transmits the received signal.

[0067] 3. Obtain wireless channel data

[0068] Actual channel data is acquired in a dedicated offline measurement environment. Backscatter tags and user terminals are deployed within a designated target area. A signal generator generates a probe signal simulating a base station transmission, with the backscatter tags in a reflective state. The user device's receiving module simultaneously collects signal characteristics and location information. The user device moves to each measurement point along a predetermined trajectory and records channel parameters such as received signal power, path loss, shadow fading, multipath components, angle of arrival, and angle of launch. Each set of measurement data is stored synchronously with its corresponding spatial location to form a standardized dataset. Repeated measurements are performed across the target area to ensure a reasonable spatial density of data. After data collection is complete, anomaly detection and cleaning are performed to remove invalid or anomalous measurement points, ultimately resulting in an offline dataset used to construct a wireless channel knowledge graph.

[0069] 4. Build an environmental model for spatial perception

[0070] First, the target area is divided into N x ×N y The grid cells of the target area are:

[0071] G={(iΔx,jΔy)∣i=1,...,N x ,j=1,...,N y}

[0072] Among them, Δx and Δy are the sizes of the grid cells in the x and y directions, which control the accuracy of the grid resolution. i and j are the indices of the grid cells in the x and y directions. The space loss field model is used to model the environment. A complete path loss model is established by comprehensively considering the effects of free space propagation and environmental shadows. Considering the effects of free space propagation and environmental shadows, a path loss model is established, and the logarithmic distance loss model is used to represent the free space loss: the reference path loss β is defined at the reference distance d0, and the influence of the propagation distance is represented by the path loss exponent α, α>0, d represents the distance from the transmitter to the receiver, and the free space path loss is expressed as:

[0073]

[0074] The shadow fading term is introduced to reflect the additional loss caused by obstacles. Based on the aforementioned discretized spatial loss field, the shadow loss is modeled as a weighted superposition of the loss contributions of each grid:

[0075]

[0076] Among them, x i and x j Represent the locations of the transmitting and receiving points respectively, φ1(x i ,x j )=|xi -x j | is the direct path distance, φ2(x p ;x i ,x j )=|x p -x i |+|x p -x j | indicates passing through the middle point x p The weight function w(φ1,φ2) is based on the normalized ellipse model and is defined as follows: if φ2≤φ1+Δ (Δ is half wavelength), then w is (or take 1 after normalization), otherwise w=0. δ p is the loss value of the p-th spatial unit (the attenuation caused by the obstacle). Thus, the shadow fading term S(x i ,x j ) is actually equivalent to the loss value δ of all grid cells passed by the signal propagation path within the first Fresnel ellipse area. p Sum. Combining the free space loss and shadow loss, we get the total path loss model for the link (expressed in dB):

[0077]

[0078] The above formula fully describes the various components of path loss: β is the path loss constant term at the reference distance d0, α is the distance loss exponent, δ p is the loss value of the pth spatial unit, and w(φ1, φ2) is the weight factor when the signal path passes through the grid unit.

[0079] 5. Gaussian Process Regression Prediction

[0080] Random shadow fading is captured through residuals, and a Gaussian process is used to model and predict it, thus introducing data-driven corrections into the model. After extracting the residuals and analyzing their distribution characteristics, it is assumed that the residual terms approximately follow a zero-mean Gaussian random field, whose covariance is determined by the distance relationship between spatial locations. Based on this assumption, a Gaussian process regression model is introduced to fit the residuals.

[0081] First, we define the input and output of the residual term and construct a channel map for the coverage area of ​​a fixed base station. The input is the receiving terminal location (x, y) (assuming the transmitting terminal is known and fixed); the output is the loss residual value r(x) at the corresponding location, which will serve as the target variable of the Gaussian process. After mean normalization, the Gaussian process establishes a mapping model f:(x, y)→r from location coordinates to residual values. An isotropic Gaussian kernel function is selected and the correlation between two locations is calculated using the Euclidean distance between them, which is the following:

[0082]

[0083] Where l is called the length scale, which determines the decay rate of the correlation between spatial positions. is the prior amplitude of the channel residual variance. When two locations are very close, that is, the distance |x p -x q When | is very small, the kernel function value is close to Indicates that the residuals are highly correlated (numerically similar). As the distance increases, the kernel function exponential term approaches zero, the correlation decreases, and the residuals of distant points are considered to be almost uncorrelated. By adjusting l, the relevant distance scale of shadow fading in the actual environment can be matched. The shadow correlation distance in cellular networks is usually tens of meters. In Gaussian process regression, assuming that the residuals obey the zero-mean Gaussian process prior, the posterior prediction is also Gaussian distributed, and the predicted mean and variance can be obtained through a closed-form solution. The predicted mean of Gaussian process regression can be expressed as:

[0084]

[0085] where r=[r(x1),...,r(x N )] T is the training residual vector, K is the N×N training sample covariance matrix, and its element K ij =k(x i ,x j ) represents the sample x i with x j The correlation between x * represents any new test position in the region, k * =[k * (x * ,x1),...,k * (x * ,x N )] T Represents the new position x * and the covariance vector between all training points, is the noise variance. * When close to some training points, these points are close to x * The correlation k * (x * ,x i ) is high, its residual value will have a larger weight in the prediction; conversely, for points with low correlation and long distance, the weight will be smaller. The prediction formula derived from the above Gaussian process can be used to estimate the path loss residual at any location based on limited measurement data.

[0086] Combining the residual correction term predicted by the Gaussian process with the previous path loss benchmark model, we can estimate the path loss for any location in the continuous space. Specifically, for any point x in the region * , and its path loss is estimated as:

[0087]

[0088] Among them, L total (x * ) represents the total path loss model of the link, is the predicted mean of the Gaussian process regression.

[0089] 6. Select a precoding scheme

[0090] The signal-to-interference-noise ratio (SINR) is used as a performance evaluation metric to assess the channel quality of each deployment point. To address the low SINR problem in links to some backscatter tags, a novel precoding scheme, hybrid MRC precoding, is chosen. The hybrid MRC precoding matrix is ​​expressed as:

[0091] v k =h k +g k

[0092] where h k represents the direct link channel matrix, g k The backscatter tag link channel matrix is ​​first decoded by the direct link signal and then the continuous interference cancellation technique is used. The formula is as follows

[0093]

[0094] Among them, K represents the number of users, y k Indicates the received signal, p i Indicates the transmit power, v k Represents the receiving combination vector, h i represents the direct link channel matrix, represents the direct link interference signal, It represents its average value, that is, the removed direct link interference signal, s i Indicates direct link information symbol, c i represents the backscatter tag link information symbol, Γ represents the reflection coefficient of the backscatter tag, N k Represents a noise signal; it removes some interference from the direct link when decoding the backscatter tag signal, thereby significantly improving the signal-to-interference-noise ratio of the backscatter tag link.

[0095] 7. Deploy backscatter tags

[0096] Compare the signal-to-interference-and-noise ratio (SINR) at various locations in the target area and prioritize deploying backscatter tags at locations with higher SINR to ensure better communication quality and reduce the impact of interference on signals.

[0097] In summary, the present invention optimizes backscatter tag deployment through wireless channel knowledge graph, provides accurate interference prediction and environmental adaptability, and provides an efficient, low-cost, and low-power solution for the future 6G passive Internet of Things.

[0098] The descriptions and practices disclosed in this invention are easy to understand and comprehend for those skilled in the art, and modifications and refinements may be made without departing from the principles of the invention. Therefore, modifications and improvements made without departing from the spirit of the invention should also be considered within the scope of protection of this invention.

Claims

1. A backscatter tag deployment method based on channel knowledge graph, characterized in that: The steps include: Step 101: Construct a MIMO communication experimental scenario including backscatter tags, and in a controlled offline measurement environment, use a signal generator to simulate base station signal transmission and a receiver to receive signals to collect actual wireless channel data for subsequent modeling and analysis. Step 102: Using the spatial loss field modeling method, the collected actual wireless channel data is used to construct a spatial perception model that describes the propagation characteristics of wireless signals. Step 103: Using the Gaussian process regression method, the shadow fading of the spatial position is continuously modeled by introducing a residual term, and the shadow loss distribution of each position is inferred, thereby constructing a wireless channel knowledge graph; Step 104: Based on the constructed wireless channel knowledge graph, taking into account the path loss model and environmental characteristics, an interference distribution map of the wireless environment is obtained. Based on the interference distribution map of the target area, the area with the least interference in the environment is analyzed, and areas with low interference power and high channel stability are selected as candidate deployment locations for backscatter tags. Step 105: Evaluate the channel quality of each candidate deployment location using the signal-to-interference-and-noise ratio (SINR) as a performance evaluation metric. Addressing the issue of low SINR values ​​for some backscatter tag links, select a precoding strategy that matches the backscatter tag link to optimize the candidate deployment location and precoding strategy. Step 106: Based on the optimized candidate deployment locations and precoding strategies, during the actual system deployment phase, the signal-to-interference-and-noise ratios of various locations in the target area are compared to evaluate the communication quality of the backscatter tag link. The deployment locations are further adjusted to complete the precise deployment of the backscatter tags.

2. A backscatter tag deployment method based on channel knowledge graph according to claim 1, characterized in that: Step 101 is specifically as follows: the MIMO communication experimental scenario includes a base station, several backscatter tags and multiple users; wherein, the base station is denoted as BS, located at a fixed position (0,0) in a two-dimensional coordinate system, and the backscatter tag set is denoted as Actual channel data is obtained in an offline measurement environment. Backscatter tags and user terminals are deployed in the set target area. A signal generator is used to generate a detection signal to simulate the transmission of the base station. The backscatter tag is in a reflective working state. The user equipment receiving module synchronously collects the received signal characteristics and location information. The user terminal moves to each measurement point according to the predetermined trajectory and records the received channel parameters including signal power, path loss, shadow fading, multipath component, arrival angle, and transmission angle. Each set of measurement data is stored synchronously with the corresponding spatial position to form a standardized measurement data pair. After repeated measurements, the target deployment area is covered. After the acquisition is completed, the data set is detected and cleaned for anomalies to remove invalid or abnormal measurement points, and finally an offline measurement data set is obtained for subsequent construction of the wireless channel knowledge graph.

3. The backscatter tag deployment method based on channel knowledge graph according to claim 1 is characterized in that: Step 102 is specifically: divide the target area into N x ×N y The size of each grid cell is Δx and Δy, and the discretized grid point set of the target area is: G={(iΔx,jΔy)∣i=1,...,N x ,j=1,...,N y } Where Δx and Δy are the sizes of the grid cells in the x and y directions, which control the accuracy of the grid resolution. i and j are the indices of the grid cells in the x and y directions. The spatial loss field model is used to model the environment. The spatial loss field model of the wireless channel uses a normalized ellipse model to describe the path loss distribution between the transmitter and the receiver. The normalized ellipse model assumes that all positions within the ellipsoid formed by the two points of transmission and reception have an impact on the shadow fading of the link; when a point is within the ellipse with the transmitter and receiver as the focus, it is considered that the point contributes to the signal propagation, otherwise the impact is negligible; for all points within the ellipse, when a position is within the ellipse, its path weight is constant Where φ1 is the direct path length, and w = 0 outside the area. Considering the influence of free space propagation and environmental shadows, a path loss model is established, and the logarithmic distance loss model is used to represent the free space loss: the reference path loss β is defined at the reference distance d0, and the influence of the propagation distance is represented by the path loss exponent α, α>0, d represents the distance from the transmitter to the receiver, and the free space path loss is expressed as: The shadow fading term is introduced to reflect the additional loss caused by obstacles. Based on the aforementioned discretized spatial loss field, the shadow loss is modeled as a weighted superposition of the loss contributions of each grid: Among them, x i and x j Represent the locations of the transmitting and receiving points respectively, φ1(x i ,x j )=|x i -x j | is the direct path distance, φ2(x p ;x i ,x j )=|x p -x i |+|x p -x j | indicates passing through the middle point x p The diffraction path length, weight function w(φ1,φ2) is based on the normalized ellipse model and is defined as: if φ2≤φ1+Δ, Δ is half wavelength, then w is Otherwise w=0; δ p is the loss value of the p-th spatial unit, which is the attenuation caused by obstacles; the shadow fading term S(x i ,x j ) is equivalent to the loss value δ of all grid cells passed by the signal propagation path within the first Fresnel ellipse area. p Sum; combining free space loss and shadow loss, we get the total path loss model of the link: Where β is the reference path loss at the reference distance d0, α is the path loss exponent, and δ p is the loss value of the pth spatial unit, and w(φ1, φ2) is the weight factor when the signal path passes through the grid unit.

4. The backscatter tag deployment method based on channel knowledge graph according to claim 1 is characterized in that: Step 103 is specifically as follows: the residual reflects random shadow fading, and a Gaussian process is used to model and predict it, thereby introducing data-driven correction into the model; after extracting the residual and analyzing its distribution characteristics, the residual term is assumed to approximately obey a zero-mean Gaussian random field, whose covariance is determined by the distance relationship between spatial positions, and a Gaussian process regression model is introduced to fit the residual. A channel map is constructed for the coverage area of ​​a fixed base station, and the position (x, y) of the receiving terminal is selected as input, and the loss residual value r(x) of the corresponding position is output; after mean normalization, the Gaussian process establishes a mapping model f:(x, y)→r from the position coordinate to the residual value; an isotropic Gaussian kernel function is selected and the correlation between the two positions is calculated using the Euclidean distance between the two positions, which is in the form of: Among them, l is called the length scale, which determines the decay rate of the correlation between spatial positions. is the prior variance of the channel residual; when the distance |x p -x q |Close to 0, the kernel function value is close to Indicates that the residuals are highly correlated; as the distance increases, the kernel function exponential term approaches zero, the correlation decreases, and the residuals of distant points are considered to be almost uncorrelated; by adjusting l to match the relevant distance scale of shadow fading in the actual environment, in Gaussian process regression, assuming that the residuals obey the zero-mean Gaussian process prior, the posterior prediction is also Gaussian distributed, and the predicted mean of Gaussian process regression is expressed as: Where r=[r(x1),...,r(x N )] T is the training residual vector, K is the N×N training sample covariance matrix, and its element K ij =k(x i ,x j ) represents the sample x i with x j The correlation between x * represents any new test position in the region, k * =[k * (x * ,x1),...,k * (x * ,x N )] T Represents the new position x * and the covariance vector between all training points, The noise variance is the residual prediction for the new position, which is the weighted average of the residual values ​​of each observation in the training set. The weight is determined by the correlation between the new point and each known point calculated by the kernel function, and is normalized by the inverse matrix of the autocorrelation of the training points. The residual correction term obtained by the Gaussian process prediction is combined with the previous path loss benchmark model to give an estimated value of the path loss for any position in the continuous space. For any point x in the region, * , and its path loss is estimated as: Among them, L total (x * ) represents the total path loss model of the link, is the predicted mean of the Gaussian process regression.

5. The backscatter tag deployment method based on channel knowledge graph according to claim 1 is characterized in that: Step 104 specifically includes: based on the constructed wireless channel knowledge graph, taking into account the path loss model and environmental characteristics, obtaining the interference distribution of the wireless environment. The path loss calculation formula is as follows: P signal (x r )=P transmit -PL(x r ) Among them, P signal (x r ) is the received signal power, P transmit is the transmission power, PL(x r ) is the distance from the transmitting point to the receiving point x r Path loss; Then calculate the interference signal power: P interf (x r ,x i )=P transmit (x i )-PL(x r ,x i ) Among them, P interf (x r ,x i ) is the interference source x i For receiving point x r The interference power, P transmit (x i ) is the interference source x i The transmission power, PL(x r ,x i ) is the interference source x i To receiving point x r Path loss; Based on the interference distribution and the strength of the target signal, the SINR at each location is calculated to assess the interference level at each location. The signal-to-interference-and-noise ratio is calculated as follows: Among them, P signal (x r ) is the received power of the target signal, is the total interference source to the receiving point x r The sum of the interference powers, P noise is the noise power; For each position x r , calculate its signal to interference and noise ratio, and obtain the interference distribution map of the target area.

6. The backscatter tag deployment method based on channel knowledge graph according to claim 1 is characterized in that: Step 105 specifically includes: evaluating the channel quality of each deployment point using the signal-to-interference-and-noise ratio as a performance evaluation indicator. To address the problem of low signal-to-interference-and-noise ratio values ​​for links to some backscatter tags, a hybrid MRC precoding matrix is ​​selected. The hybrid MRC precoding matrix is ​​expressed as: v k =h k +g k Among them, h k represents the direct link channel matrix, g k The backscatter tag link channel matrix is ​​represented by first decoding the direct link signal and then using the successive interference cancellation technique. The formula is as follows: Among them, K represents the number of users, y k Indicates the received signal, p i Indicates the transmit power, v k Represents the received combination vector, h i represents the direct link channel matrix, represents the direct link interference signal, It represents its average value, that is, the removed direct link interference signal, s i Indicates direct link information symbol, c i represents the backscatter tag link information symbol, Γ represents the reflection coefficient of the backscatter tag, N k Represents a noise signal; it removes some interference from the direct link when decoding the backscatter tag signal, improving the signal-to-interference-noise ratio of the backscatter tag link.

7. The backscatter tag deployment method based on channel knowledge graph according to claim 1 is characterized in that: Step 106 : Compare the signal-to-interference-noise ratios (SINRs) of various locations in the target area, and prioritize deploying backscatter tags at locations with higher SINRs.

8. A computer device comprising a memory, a processor, and a computer program stored in the memory, wherein: The processor executes the computer program to implement the steps of the method according to claim 1.

9. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by a processor, the steps of the method according to claim 1 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method according to claim 1 are implemented.

Images