Method for channel knowledge map based wireless communication environment change perception

Abstract

The application discloses a wireless communication environment change perception method based on a channel knowledge map, relates to the technical field of perception, and acquires the position and channel information of users in a local area and calls the position and channel information of multiple grid points; the actual departure direction and the actual arrival direction are calculated according to the path information of each reflection path of the user, a maximum likelihood estimation problem is constructed, an alternating iterative optimization is adopted to obtain the optimal departure direction and the optimal arrival direction, and the initial estimated position of the reflection point of each reflection path of the user is calculated; based on the mirror consistency principle, a clustering correction algorithm is adopted to determine real-time mirror base station classes and iteratively correct the initial estimated position of the reflection point to obtain the estimated position of the reflection point in the real-time mirror base station class; the position and channel information of the grid points are processed in the same way to obtain historical mirror base station classes, the real-time mirror base station classes and the historical mirror base station classes are matched based on the minimum matching principle, and the reflection surface change condition is tested and judged, so that efficient and high-precision environment perception is realized.

Description

Technical Field

[0001] This invention relates to the field of sensing technology, and more specifically to a method for sensing changes in the wireless communication environment based on a channel knowledge map. Background Technology

[0002] To meet the ultra-high speed, low latency, wide connectivity, and intelligent communication requirements of 5G and 6G, and to compensate for the information transmission distortion caused by the stringent requirements of millimeter waves on communication quality and environmental adaptability, it is necessary to utilize intelligent sensing technology to test and verify changes in the wireless communication environment in real time based on wireless communication information, accurately obtain interference distribution, help dynamically optimize communication parameters, improve communication quality and reliability, and achieve the goal of integrated communication and sensing.

[0003] In existing wireless environment change sensing technologies, when the number of reflection paths at the same location changes significantly at different times, there are mature algorithms to sense the environment change. When the number of reflection paths at the same location does not change significantly at different times, existing technologies mostly rely on direct comparison of channel information at different times to assist in environment sensing. However, channel information is complex in dimensions, which affects the operating efficiency of sensing algorithms, and the acquisition of channel information has unavoidable noise errors, which can easily lead to misjudgment of environmental changes. Summary of the Invention

[0004] This invention addresses the shortcomings of existing technologies by proposing a wireless communication environment change perception method based on channel knowledge maps, in order to achieve efficient and high-precision wireless environment change perception.

[0005] The technical solution to achieve the purpose of this invention is as follows:

[0006] A method for sensing changes in the wireless communication environment based on channel knowledge maps includes the following steps:

[0007] From base stations in local areas Obtain location and channel information from the communication process of each user to construct a user information set. Retrieve local area from channel knowledge map Based on the location and channel information of each grid point, a map information set is constructed. ;

[0008] Using a single-path optimization algorithm, starting from the first path... Channel information for individual users Extract the first Path information of the reflection path And calculate the actual departure direction. and actual arrival direction The maximum likelihood estimation problem is constructed with the goal of minimizing noise error, and a penalty function is introduced to handle coplanar constraints. Construct a multiconvex problem and use alternating iterative optimization to obtain the i-th... The first user's Optimal departure direction of the reflection path and optimal arrival direction Reference base station location With the Location of individual users Calculate the first The first user's Preliminary estimated location of reflection point along the reflection path , , This represents the number of reflection paths.

[0009] Based on the mirror consistency principle, a clustering correction algorithm is used to calculate in each iteration. The real-time mirror base station location for each user's reflection path is determined through clustering. Each real-time mirror base station class and its corresponding real-time centroid location, based on the first Real-time centroid location of a real-time mirror base station class Amendment No. The initial estimated position of the reflection point for each reflection path in each real-time mirror base station class is calculated, and the process is repeated iteratively until convergence is achieved, yielding the first... Estimated location of reflection point for each reflection path in each real-time mirror base station class , This is the total number of reflecting surfaces that is the same as the number of reflection paths.

[0010] Map information sets are processed using single-path optimization and clustering correction algorithms. get The estimated locations of reflection points for each historical mirror base station class and each reflection path within that class are determined based on the minimum matching principle. Real-time mirror base station class and Each historical mirror base station class is matched one by one, and the changes in the reflective surface are determined.

[0011] Furthermore, the channel knowledge map divides the global region into a grid, recording the location of each grid point and the information used to determine the channel matrix. Channel information, based on a millimeter-wave channel model, includes... The path gain, pitch angle of arrival, horizontal angle of arrival, pitch departure angle, and horizontal departure angle of each reflection path are calculated. The pitch angle of arrival and horizontal angle of arrival are two-dimensional angles when the reflection path reaches the receiving device, and the pitch departure angle and horizontal departure angle are two-dimensional angles when the reflection path is transmitted from the base station. Channel information of 30% to 50% of the grid points is collected through pilot-based channel estimation. The channel information of the remaining grid points is calculated through expectation-maximization algorithm or neural network technology, and a channel knowledge map is established.

[0012] Specifically, the single-path optimization algorithm starts from the first... Channel information for individual users Extract the first Path information of the reflection path , including the The first user's Pitch arrival angle of the reflection path Horizontal angle of arrival Pitch departure angle and horizontal departure angle Calculation based on solid geometry The first user's The actual departure direction of the reflection path and actual arrival direction Precise departure direction and precise arrival direction Satisfy coplanar constraints Sum of constraints Coplanar constraints Precise departure direction required Precise arrival direction With the The locations of users and base stations are coplanar, subject to module constraints. Precise departure direction required Precise arrival direction The modulus is 1, that is Consider maximum likelihood estimation to minimize the departure direction. , destination actual departure direction Actual destination direction The error is the optimization objective, combined with coplanar constraints. Sum of constraints Construct the maximum likelihood estimation problem.

[0013] Furthermore, coplanar constraints The Middle The first user's The direction of departure of the reflection path and direction of arrival The coupling causes the maximum likelihood estimation problem to be nonconvex, which can be addressed by using a penalty function. Handling coplanar constraints Then, by adding the objective of the maximum likelihood estimation problem, a multiconvex problem is constructed, as follows:

[0014] ,

[0015] in, Regarding the direction of departure and direction of arrival A multiconvex function.

[0016] Furthermore, the alternating iterative optimization method is used to obtain the first... The first user's Optimal departure direction of the reflection path and optimal arrival direction This includes the following steps:

[0017] Will leave the direction and direction of arrival As the departure direction in round 0 and direction of arrival Initialize the maximum number of rounds. First convergence threshold Step size reduction ratio The first round of departure optimization step size And the first round of optimization step size ;

[0018] In the In the round of iteration, , fixed number The direction the wheel arrives Substituting this into a multiconvex problem simplifies the generation of the first subproblem, while also incorporating modulo constraints. Simplified to departure direction The modulus is 1, and the first convex function in the first subproblem is solved using the Riemannian manifold method. First Euclidean gradient And projected onto the tangent space of the unit modulus to obtain the first The first Riemann gradient of the wheel According to Article Wheel departure optimization step size Amendment No. The direction of the wheel's departure Reprojection yields the first The direction of the wheel's departure Based on the line search criterion and step size limitation ratio Determine the first Wheel departure optimization step size ;

[0019] Fixed number The direction of the wheel's departure Substituting this into a multiconvex problem simplifies the generation of the second subproblem, while also incorporating modulus constraints. Simplified to direction of arrival The modulus is 1, and the Riemannian manifold method is used to correct the first... The direction the wheel arrives Reprojection yields the first The direction the wheel arrives And determine the first Wheel to arrive optimization step ;

[0020] The first Wheel of the direction of departure And the arrival direction Substitute the multi-convex function Get the first Wheel of the multi-convex function value , determine the first Wheel of the multi-convex function value The absolute value of the change is less than the first convergence threshold ;

[0021] If less than the first convergence threshold , stop iteration and the first Wheel of the direction of departure And the arrival direction As the first User of the first Optimal direction of departure of the reflection path And the optimal arrival direction , if greater than or equal to the first convergence threshold , determine whether the current round Equal to the maximum number of rounds ;

[0022] If equal to the maximum number of rounds , stop iteration and the first Wheel of the direction of departure And the arrival direction As the first User of the first Optimal direction of departure of the reflection path And the optimal arrival direction , if less than the maximum number of rounds , start the first Round of iteration.

[0023] Further, determine the first Wheel of the departure optimization step And the arrival optimization step , including the following steps:

[0024] The first Wheel of the direction of departure And the arrival direction Substitute the first convex function And the second convex function , get the first Wheel of the first convex function value And the second convex function value ;

[0025] The first The first convex function value of the wheel Add step size reduction ratio , No. Wheel departure optimization step size With the The first Riemann gradient of the wheel The product of yields the first The first convex threshold of the wheel ;

[0026] The first The second convex function value of the wheel Add step size reduction ratio , No. Wheel arrival optimization step size With the The second Riemann gradient of the wheel The product of yields the first The second convex threshold of the wheel ;

[0027] Based on the line search criterion, based on the first The first convex function value of the wheel Is it greater than the first? The first convex threshold of the wheel Choose the first Wheel departure optimization step size Set step size limit ratio With the Wheel departure optimization step size The product or the extension of the first Wheel departure optimization step size ;

[0028] Based on the The second convex function value of the wheel Is it greater than the first? The second convex threshold of the wheel Choose the first Wheel arrival optimization step size Set step size limit ratio With the Wheel arrival optimization step size The product or the extension of the first Wheel arrival optimization step size .

[0029] Specifically, the solution is obtained through alternating iterative optimization. The first user's Optimal departure direction of the reflection path and optimal arrival direction From the first Location of individual users Along the optimal arrival direction Make a ray, from the base station location along the optimal departure direction Draw two rays; the intersection of the two rays is the first ray. The first user's The reflection points along the reflection paths are determined by establishing expressions for the two rays based on symmetric equations and solving a system of equations to obtain the initial estimated positions of the reflection points. .

[0030] Specifically, based on the principle of mirror consistency, the results are calculated using the precise arrival and departure directions. The user in the first The real-time mirror base station locations of the reflection paths of each reflector coincide, calculated based on the optimal departure and arrival directions. The user in the first The real-time mirror base station locations of the reflection paths of the first reflecting surface are distributed within a local neighborhood, constituting the first... The real-time mirror base station class, for the first The first user's The reflection path, calculate the first one. Location of individual users Preliminary position of the reflection point Reflection distance between Base station location Preliminary position of the reflection point Incident distance between Summing up yields the first... The first user's Propagation distance of a reflection path , No. The first user's Real-time mirror base station location of the reflection path equal to the Location of individual users Add optimal arrival direction With transmission distance The product can be calculated similarly. Real-time mirror base station locations for each reflection path of each user are used to construct a real-time mirror base station location set. .

[0031] Furthermore, the clustering correction algorithm includes the following steps:

[0032] Set the maximum number of clustering rounds. Second convergence threshold and update ratio Real-time mirrored base station location set As the cluster set of round 0 ,Will The optimal arrival and departure directions of each user's reflection path are used as the direction set for round 0. ;

[0033] In the In the round of iteration, a clustering algorithm is used to classify the th... wheel cluster set Perform clustering to obtain the first Wheel The first real-time mirror base station class and the first The set of the center of mass of the wheel , For the first The first round The centroid position of a real-time mirror base station class;

[0034] For the The first real-time mirror base station class, based on the first The first round Centroid position of a real-time mirror base station class Calculate each user's position in the first quarter. The first reflecting surface The initial arrival direction of the wheel, according to the first... The first round Centroid position of a real-time mirror base station class With base station location Calculate the first The first reflecting surface The normal vector of the wheel is calculated based on geometric relationships to determine the number of times each user is in the first cycle. The first reflecting surface The initial departure direction of the wheel yields the first... Initial direction set of the wheel , ;

[0035] Based on the update ratio The first Initial direction set of the wheel With the Wheel direction set Weighted summation yields the first Wheel direction set ,calculate In each of the real-time mirror base station classes, each reflection path is in the... Estimate the position of the reflection point of the wheel and deduce the corresponding first... wheel cluster set ;

[0036] Calculate the first wheel cluster set With the wheel cluster set The average distance error of all real-time mirror base station locations is calculated, and it is determined whether it is less than the second convergence threshold. ;

[0037] If it is less than the second convergence threshold Stop iteration and output In each of the real-time mirror base station classes, each reflection path is in the... The estimated position of the wheel's reflection point, if greater than or equal to the second convergence threshold... Then determine the current round number. Is it equal to the maximum number of clustering rounds? ;

[0038] If equal to the maximum number of clustering rounds Stop iteration and output In each of the real-time mirror base station classes, each reflection path is in the... If the estimated position of the reflection point of the wheel is less than the maximum number of clustered wheels... , start the Round iteration.

[0039] Specifically, calculate the first The estimated position of the reflection point for each reflection path in each real-time mirror base station class is compared with the first... Hasdorf distance of the estimated location of the reflection point for each reflection path in each historical mirror base station class Construct the Hasdorf distance matrix The Hasdorf distance matrix The first line Each column is used as a first-level node, and the Hasdorf distance matrix is ​​explored sequentially based on the principle of no repeated columns. Each line, in exploring the first When the line is drawn, for the first line... Each row Level node, excluding rows 1 to 1 Rows of selected columns, based on unselected columns Each column is for Level node generation indivual Level nodes, until the exploration reaches the first level. Okay, the forest construction is complete.

[0040] Furthermore, based on the minimum matching principle, the optimal path with the smallest total distance to Hasdorf in the forest is selected, and the number of columns corresponding to each level node in the optimal path is determined. Real-time mirror base station class and The matching relationship of each historical mirror base station class is used to query the Hasdorf distance matrix by combining the number of rows and corresponding columns of each level node. obtaining a corresponding Hausdorff distance, and comparing the distance with a distance threshold comparing to decide whether the reflecting surface corresponding to each matched historical mirror base station class and real-time mirror base station class changes.

[0041] Compared with the prior art, the single-path optimization algorithm is used to extract the path information of each reflection path from the channel information of each user and calculate the actual departure direction and actual arrival direction, a maximum likelihood estimation problem is constructed with the target of minimizing noise error, a penalty function is introduced to handle the coplanar constraint to construct a multi-convex problem, and an alternating iterative optimization is adopted to obtain the optimal departure direction and optimal arrival direction of each reflection path of each user, the corresponding reflection point initial position is calculated by referring to the base station position and the position of each user, and the high-dimensional channel information of multiple paths of each user is effectively reduced and represented through the reflection point initial position.

[0042] Based on the mirror consistency principle, a clustering correction algorithm is adopted to calculate the real-time mirror base station position of each reflection path of each user in each iteration, determine all real-time mirror base station classes and corresponding real-time centroid positions through clustering, correct the reflection point initial position of each reflection path in each real-time mirror base station class based on the real-time centroid position of each real-time mirror base station class, and repeat the iteration until convergence, so as to obtain the reflection point estimation position of each reflection path in each real-time mirror base station class and weaken the influence of the collected channel information error on the reflection point position estimation.

[0043] All historical mirror base station classes and the reflection point estimation position of each reflection path in each historical mirror base station class are obtained by processing the map information set through the single-path optimization algorithm and the clustering correction algorithm, all real-time mirror base station classes are matched with all historical mirror base station classes based on the minimum matching principle, and the reflection surface change is judged, so as to realize high-precision reflection surface change perception in a wireless communication environment. BRIEF DESCRIPTION OF DRAWINGS

[0044] Figure 1 A flowchart of a wireless communication environment change perception method based on a channel knowledge map;

[0045] Figure 2 A flowchart of an alternating iterative optimization;

[0046] Figure 3 A schematic diagram of a mirror consistency principle;

[0047] Figure 4 A flowchart of a clustering correction algorithm. DETAILED DESCRIPTION

[0048] The application will be further described in detail below in combination with the drawings and embodiments.

[0049] As Figure 1As shown, a specific embodiment of the present invention discloses a method for sensing changes in the wireless communication environment based on a channel knowledge map, comprising the following steps:

[0050] GPS positioning and pilot-based channel estimation are used to obtain base station and... The location and channel information of each user in the real-time communication process are used to construct a user information set. Retrieve local area from channel knowledge map Based on the location and channel information of each grid point, a map information set is constructed. Among these, GPS positioning and pilot-based channel estimation are existing technologies and will not be discussed in detail. and The first Location and channel information of each user and The first The channel knowledge map pre-records the location and channel information of different grid points in the global area in a grid format. Total number of users This represents the total number of grid points.

[0051] Using a single-path optimization algorithm, starting from the first path... Channel information for individual users Extract the first Path information of the reflection path And calculate the actual departure direction. and actual arrival direction The maximum likelihood estimation problem is constructed with the goal of minimizing noise error, and a penalty function is introduced to handle the departure direction. , destination coplanar constraints Construct a multiconvex problem and use alternating iterative optimization to obtain the i-th... The first user's Optimal departure direction of the reflection path and optimal arrival direction Reference base station location With the Location of individual users Calculate the first The first user's Preliminary estimated location of reflection point along the reflection path , , This represents the number of reflection paths.

[0052] Based on the mirror consistency principle, a clustering correction algorithm is adopted. In each iteration, the following calculations are performed: users The real-time mirror base station locations corresponding to each reflection path are determined through clustering. Each real-time mirror base station class and its corresponding real-time centroid location, based on the first Real-time centroid location of a real-time mirror base station class Amendment No. In each real-time mirror base station class The initial position of the reflection point of the reflection path is estimated, and the iteration is repeated until convergence is obtained to obtain the first reflection point. In each real-time mirror base station class Estimated location of reflection point along a reflection path. , The total number of reflecting surfaces and the number of reflecting paths. They are identical, and each real-time mirror base station class uniquely corresponds to a reflective surface in a wireless communication environment;

[0053] Using user information set Using the same processing method, the map information set is processed through the single-path optimization algorithm and the clustering correction algorithm. get Each historical mirror base station class and within each historical mirror base station class The estimated location of the reflection point along the reflection path is determined based on the minimum matching principle. Real-time mirror base station class and Each historical mirror base station is matched one by one. Based on the matching results, it is tested whether the reflective surface in the wireless communication environment of the local area has changed. Each reflection path uniquely corresponds to a grid point.

[0054] Furthermore, the channel knowledge map divides the global region into a grid, recording the location of each grid point and the information used to determine the channel matrix. Channel information, based on a millimeter-wave channel model, includes... The path gain, elevation angle of arrival, horizontal angle of arrival, elevation departure angle, and horizontal departure angle of the reflection path are given. The elevation angle of arrival and horizontal angle of arrival are the two-dimensional angles when the reflection path reaches the receiving device, and the elevation departure angle and horizontal departure angle are the two-dimensional angles when the reflection path is transmitted from the base station. The millimeter-wave channel model is as follows:

[0055] ,

[0056] in, and These represent the number of base station antennas and the number of receiving device antennas, respectively. , , , and The first The path gain, pitch angle of arrival, horizontal angle of arrival, pitch departure angle, and horizontal departure angle of the reflection path. and These are the array response vectors of the receiving device and the base station, respectively, and are directly related to the antenna array arrangement of the receiving device and the base station. As the conjugate transpose of a vector or matrix, this embodiment uses a uniform planar antenna array. Due to the large global coverage area, the cost of collecting channel information for each grid point is too high. Therefore, a channel knowledge map is established by combining offline acquisition and online reconstruction. Channel information for 30% to 50% of the grid points is collected through pilot-based channel estimation. The channel information for the remaining grid points is calculated using the expectation-maximization algorithm or neural network technology, and a channel knowledge map is established. The method for establishing the channel knowledge map is a mature existing technology, and this application will not elaborate on it further.

[0057] Specifically, the single-path optimization algorithm starts from the first... Channel information for individual users Extract the first Path information of the reflection path ,in, , , , and The first The first user's The path gain, pitch angle of arrival, horizontal angle of arrival, pitch departure angle, and horizontal departure angle of the reflection path are based on solid geometry. The first user's The actual departure direction of the reflection path and actual arrival direction Specifically as follows:

[0058] , ,

[0059] Leaving direction and direction of arrival For error-free, precise departure direction and precise arrival direction At that time, from the first Location of individual users Along the precise direction of arrival Make a ray, from the base station location Along the precise departure direction Draw two rays; the intersection of the two rays is the first ray. The first user's Reflection point location of a reflection path At this point, the precise departure direction and precise arrival direction Strictly satisfy coplanar constraints Sum of constraints Coplanar constraints Precise departure direction required Precise arrival direction With the The locations of users and base stations are coplanar, subject to module constraints. Precise departure direction required Precise arrival direction The modulus is 1, that is Because pilot-based channel estimation suffers from slight noise errors due to noise interference, it is based on the actual departure direction. and actual arrival direction Two defined rays may not intersect; consider maximum likelihood estimation to pinpoint the exact departure direction. Precise arrival direction actual departure direction Actual destination direction Very close and satisfy coplanar constraints Sum of constraints Therefore, to find the coplanarity constraint Sum of constraints And the actual departure direction Actual destination direction The departure direction with the smallest error , destination To achieve this goal, we construct a maximum likelihood estimation problem, as follows:

[0060] ,

[0061] The optimal departure direction is obtained by solving the maximum likelihood estimation problem. and optimal arrival direction It may not be completely error-free, but compared to the actual departure direction... and actual arrival direction More closely approximating the precise departure direction and precise arrival direction .

[0062] Furthermore, due to coplanar constraints The Middle The first user's The direction of departure of the reflection path and direction of arrival The mutual coupling leads to the non-convexity of the maximum likelihood estimation problem, which can be addressed by utilizing a penalty function. Handling coplanar constraints penalty function Specifically as follows:

[0063] ,

[0064] in, As the weight of the penalty function, the penalty function Adding the optimization objective of the maximum likelihood estimation problem to construct a multiconvex problem, as follows:

[0065] ,

[0066] in, Regarding the direction of departure and direction of arrival A multiconvex function.

[0067] like Figure 2 As shown, further, the alternating iterative optimization method is used to obtain the th... The first user's Optimal departure direction of the reflection path and optimal arrival direction This includes the following steps:

[0068] Will leave the direction and direction of arrival These serve as the departure directions for round 0. and direction of arrival Initialize the maximum number of rounds. First convergence threshold Step size reduction ratio The first round of departure optimization step size And the first round of optimization step size Among them, the step size reduction ratio ;

[0069] In the In the round of iteration, , fixed number The direction the wheel arrives The multiconvex problem is simplified into a first subproblem, as follows:

[0070] ,

[0071] in, It is the first convex function. For the first The direction the wheel arrives Substitute penalty function The obtained number The arrival product matrix of the rounds, For the first The direction the wheel arrives and the actual direction of arrival The error, For model constraints The first sub-constraint in ;

[0072] Solving the first convex function using the Riemannian manifold method First Euclidean gradient Projecting onto the tangent space of the unit modulus yields the first... The first Riemann gradient of the wheel And according to the first Wheel departure optimization step size Amendment No. The direction of the wheel's departure Reprojection yields the first The direction of the wheel's departure And based on the line search criterion, the decision is made according to the step size limitation ratio. Adjustment of the Wheel departure optimization step size Or continue to use it, to obtain the first Wheel departure optimization step size ;

[0073] Fixed number The direction of the wheel's departure The multiconvex problem is simplified into a second subproblem, as follows:

[0074] ,

[0075] in, It is the second convex function. For the first The direction of the wheel's departure Substitute penalty function The obtained number The wheel leaves the product matrix. For the first The direction of the wheel's departure actual departure direction The error, For model constraints The second sub-constraint ;

[0076] Solving the second convex function using the Riemannian manifold method The second Euclidean gradient Projecting onto the tangent space of the unit modulus yields the first... The second Riemann gradient of the wheel And according to the first Wheel arrival optimization step size Amendment No. The direction the wheel arrives Reprojection yields the first The direction the wheel arrives And based on the line search criterion, the decision is made according to the step size limitation ratio. Adjustment of the Wheel arrival optimization step size Or continue to use it, to obtain the first Wheel arrival optimization step size ;

[0077] The first The direction of the wheel's departure and direction of arrival Substitute the multiconvex function Get the first The multiconvex function value of the wheel Judgment and the first The multiconvex function value of the wheel Is the absolute value of the change less than the first convergence threshold? ;

[0078] If it is less than the first convergence threshold Stop iteration and the first The direction of the wheel's departure and direction of arrival As the first The first user's Optimal departure direction of the reflection path and optimal arrival direction Output, if greater than or equal to the first convergence threshold Then determine the current round number. Is it equal to the maximum number of rounds? ;

[0079] If equal to the maximum number of rounds Stop iteration and the first The direction of the wheel's departure and direction of arrival As the first The first user's Optimal departure direction of the reflection path and optimal arrival direction Output, if less than the maximum number of rounds , start the Round iteration.

[0080] Furthermore, the Riemannian manifold method is used to handle the first sub-constraints of the first subproblem. Second sub-constraints with the second subproblem In the The round of iteration includes the following steps:

[0081] For the first convex function Regarding the first The user The direction of departure of the path Solve for the first Euclidean gradient For the second convex function Regarding the first The user The direction of arrival of the path Solving for the second Euclidean gradient The details are as follows:

[0082] ,

[0083] ,

[0084] in, Transpose of a vector or matrix;

[0085] To satisfy the first sub-constraint With second sub-constraint The first Euclidean gradient With the second Euclidean gradient Projecting these onto the tangent space of the unit modulus manifold, we obtain the first... The first Riemann gradient of the wheel Second Riemann gradient The details are as follows:

[0086] ,

[0087] ,

[0088] Among them, the unit modulus manifold refers to a surface with a modulus of 1. Updating the tangent space of the unit modulus manifold according to the Riemann gradient can always ensure that the constraint of a modulus of 1 is satisfied.

[0089] According to the The first Riemann gradient of the wheel With the Wheel departure optimization step size Amendment No. The direction of the wheel's departure The images are then projected back to the original space, according to the first... The second Riemann gradient of the wheel With the Wheel arrival optimization step size Amendment No. The direction the wheel arrives The images are then projected back into the original space, yielding the first two images respectively. The direction of the wheel's departure and direction of arrival The details are as follows:

[0090] ,

[0091] ;

[0092] the first convex function value of the first round is obtained by substituting the leaving direction of the first round into the first convex function and the arriving direction of the first round into the second convex function ; the second convex function value of the first round is obtained by substituting the leaving direction of the first round into the first convex function and the arriving direction of the first round into the second convex function ;the first convex threshold value of the first round is obtained by adding the step reduction ratio, the leaving optimized step of the first round and the product of the first Riemann gradient of the first round to the first convex function value of the first round ;the second convex threshold value of the first round is obtained by adding the step reduction ratio, the arriving optimized step of the first round and the product of the second Riemann gradient of the first round to the second convex function value of the first round ; ; ; ; ; ; ;

[0093] whether the first convex function value of the first round is greater than the first convex threshold value of the first round is determined based on the line search criterion, if greater than the first convex threshold value of the first round, the leaving optimized step of the first round is set as the product of the step reduction ratio and the leaving optimized step of the first round, if less than or equal to the first convex threshold value of the first round, the leaving optimized step of the first round is set as the leaving optimized step of the first round ; ;

[0094] whether the second convex function value of the first round is greater than the second convex threshold value of the first round is determined based on the line search criterion, if greater than the second convex threshold value of the first round, the arriving optimized step of the first round is set as the product of the step reduction ratio and the arriving optimized step of the first round, if less than or equal to the second convex threshold value of the first round, the arriving optimized step of the first round is set as the arriving optimized step of the first round ;​​​​​​​​​​​​​​​​​​​​​​​​​​ Is it greater than the first? The second convex threshold of the wheel If greater than the first The second convex threshold of the wheel Then the first Wheel arrival optimization step size Set step size limit ratio With the Wheel arrival optimization step size If the product of the first and second products is less than or equal to the first product, then the product of the first and second products is less than The second convex threshold of the wheel Then the first Wheel arrival optimization step size Set as number Wheel arrival optimization step size .

[0095] Specifically, the solution is obtained through alternating iterative optimization. The first user's Optimal departure direction of the reflection path and optimal arrival direction From the first Location of individual users Along the optimal arrival direction Make a ray, from the base station location along the optimal departure direction Draw two rays; the intersection of the two rays is the first ray. The first user's The reflection points along the reflection paths are determined by establishing expressions for the two rays based on symmetric equations and solving a system of equations to obtain the initial estimated positions of the reflection points. .

[0096] like Figure 3 As shown, specifically, based on the principle of mirror consistency, under the premise of accurate arrival and departure directions without error, The user in the first The real-time mirror base station locations corresponding to the reflection paths of each reflector completely overlap, and the real-time mirror base station locations are identical to the base station locations. The line connecting the two is perpendicular to the first. One reflective surface;

[0097] because The optimal departure and arrival directions for each user's reflection path still have errors, therefore... The user in the first The real-time mirror base station locations corresponding to the reflection paths of each reflector are distributed within a local neighborhood. From a spatial perspective, The user in the first The real-time mirror base station position corresponding to the reflection path of the first reflection surface forms a first real-time mirror base station class

[0098] For the first reflection path of the first user, the reflection distance between the position of the first user and the initial estimation position of the reflection point The reflection distance between the position of the first user and the initial estimation position of the reflection point The reflection distance between the position of the first user and the initial estimation position of the reflection point The reflection distance between the position of the first user and the initial estimation position of the reflection point The reflection distance between the position of the first user and the initial estimation position of the reflection point The reflection distance between the position of the first user and the initial estimation position of the reflection point The reflection distance between the position of the first user and the initial estimation position of the reflection point The reflection distance between the position of the first user and the initial estimation position of the reflection point The reflection distance between the position of the first user and the initial estimation position of the reflection point The reflection distance between the position of the first user and the initial estimation position of the reflection point The reflection distance between the position of the first user and the initial estimation position of the reflection point The reflection distance between the position of the first user and the initial estimation position of the reflection point The reflection distance between the position of the first user and the initial estimation position of the reflection point The reflection distance between the position of the first user and the initial estimation position of the reflection point The reflection distance between the position of the first user and the initial estimation position of the reflection point The reflection distance between the position of the first user and the initial estimation position of the reflection point The reflection distance between the position of the first user and the initial estimation position of the reflection point The reflection distance between the position of the first user and the initial estimation position of the reflection point The reflection distance between the position of the first user and the initial estimation position of the reflection point The reflection distance between the position of the first user and the initial estimation position of the reflection point The reflection distance between the position of the first user and the initial estimation position of the reflection point The reflection distance between the position of the first user and the initial estimation position of the reflection point The reflection distance between the position of the first user and the initial estimation position of the reflection point The reflection distance between the position of the first user and the initial estimation position of the reflection point The reflection distance between the position of the first user and the initial estimation position of the reflection point The reflection distance between the position of the first user and the initial estimation position of the reflection point The reflection distance between the position of the first user and the initial estimation position of the reflection point The reflection distance between the position of the first user and the initial estimation position of the reflection point The reflection distance between the position of the first user and the initial estimation position of the reflection point The reflection distance between the position of the first user and the initial estimation position of the reflection point The reflection distance between the position of the first user and the initial estimation position of the reflection point The reflection distance between the position of the first user and the initial estimation position of the reflection point The reflection distance between the position of the first user and the initial estimation position of the reflection point The reflection distance between the position of the first user and the initial estimation position of the reflection point The reflection distance between the position of the first user and the initial estimation position of the reflection point

[0099] As shown in FIG. 8, further, the clustering correction algorithm comprises the following steps: Figure 4 Setting a maximum clustering round number, a second convergence threshold and an update ratio

[0100] Setting a maximum clustering round number, a second convergence threshold and an update ratio Setting a maximum clustering round number, a second convergence threshold and an update ratio Setting a maximum clustering round number, a second convergence threshold and an update ratio Setting a maximum clustering round number, a second convergence threshold and an update ratio​​​​ the cluster set of the 0th round , the optimal arrival direction and the optimal departure direction of each reflection path of the user obtained by the iterative optimization solution of the 0th round , update the proportion ;

[0101] In the iteration of the 0th round, the cluster algorithm is used to cluster the cluster set of the 0th round , to obtain the cluster set of the 0th round , the cluster algorithm includes K-means, K-means++, hierarchical clustering and density clustering, K-means and K-means++ need to set the number of clusters equal to the total number of reflecting surfaces in advance , the termination condition of hierarchical clustering and density clustering is set to the number of clusters equal to the total number of reflecting surfaces , the centroid position of the 0th round , the centroid position of the 0th round , the centroid position of the 0th round , the centroid position of the 0th round , the centroid position of the 0th round , the centroid position of the 0th round , the centroid position of the 0th round , the centroid position of the 0th round

[0102] For the 0th round , the initial arrival direction of each user at the 0th round , the centroid position of the 0th round , the centroid position of the 0th round , the initial arrival direction of each user at the 0th round , the normal vector of the 0th round , the normal vector of the 0th round , the normal vector of the 0th round , the normal vector of the 0th round , the normal vector of the 0th round , the normal vector of the 0th round , the normal vector of the 0th round , the normal vector of the 0th round , the normal vector of the 0th round , the normal vector of the 0th round , the normal vector of the 0th round , the normal vector of the 0th round , the normal vector of the 0th round , the normal vector of the 0th round , the normal vector of the 0th round , the normal vector of the 0th round , the normal vector of the 0th round The initial departure direction of the wheel yields the first... Initial direction set of the wheel , ;

[0103] Based on the update ratio The first Initial direction set of the wheel With the Wheel direction set Weighted summation yields the first Wheel direction set According to Article Wheel direction set calculate In each of the real-time mirror base station classes, each reflection path is in the... The position of the reflection point of the wheel is estimated and the corresponding first reflection point is calculated simultaneously. wheel cluster set ;

[0104] Calculate the first wheel cluster set With the wheel cluster set middle The average distance error of each real-time mirror base station location is used to determine whether it is less than the second convergence threshold. ;

[0105] If it is less than the second convergence threshold Stop iteration and output In each of the real-time mirror base station classes, each reflection path is in the... The estimated position of the wheel's reflection point, if greater than or equal to the second convergence threshold... Then determine the current round number. Is it equal to the maximum number of clustering rounds? ;

[0106] If equal to the maximum number of clustering rounds Stop iteration and output In each of the real-time mirror base station classes, each reflection path is in the... If the estimated position of the reflection point of the wheel is less than the maximum number of clustered wheels... , start the Round iteration.

[0107] Specifically, calculate the first In each real-time mirror base station class The estimated location of the reflection point of the reflection path is the same as that of the first reflection path. In the category of historical mirror base stations Hasdorf distance of the estimated location of the reflection point on the reflection path , the Hausdorff distance is a commonly used index for measuring the similarity of sets, and the Hausdorff distance is smaller, the higher the similarity between the th real-time mirror base station class and the th historical mirror base station class, the Hausdorff distance between the th real-time mirror base station class and the th historical mirror base station class is solved respectively, a Hausdorff distance matrix with a dimension of is constructed , the Hausdorff distance matrix , the Hausdorff distance between the th real-time mirror base station class and the th historical mirror base station class is , the first row of the Hausdorff distance matrix , the first row of the Hausdorff distance matrix , the first row of the Hausdorff distance matrix , the first row of the Hausdorff distance matrix , each column of the first row of the Hausdorff distance matrix , when the th row is explored, for each level node of the th row, the path formed by the node is backtracked to the corresponding first level node to determine the selected columns of the previous rows, and level nodes are generated for each level node based on the remaining unselected columns to ensure that the real-time mirror base station class and the historical mirror base station class are always one-to-one correspondence, until the th row of the Hausdorff distance matrix is explored, at this time each level node only has a unique unselected column, and the distance forest construction is completed.

[0108] Further, based on the minimum matching principle, the optimal path with the smallest total Hausdorff distance in the distance forest is selected to determine the matching relationship between the th real-time mirror base station class and the th historical mirror base station class, for example, the level node in the optimal path corresponds to the th column, indicating that the th real-time mirror base station class and the th historical mirror base station class have the highest probability of corresponding to the same reflection surface for the reflection point, and the Hausdorff distance in the th row and the th column of the Hausdorff distance matrix is queried, and the Hausdorff distance Is it less than the distance threshold? If it is less than the distance threshold This indicates the first The first real-time mirror base station class and the first Each historical mirror base station class corresponds to the same reflective surface, and between the recording time of the channel knowledge map and the user's collection time, the first... The first real-time mirror base station class and the first If the reflective surface corresponding to a historical mirror base station class has not changed, and is greater than or equal to the distance threshold... This indicates the first The first real-time mirror base station class and the first Each historical mirror base station class corresponds to a different reflector surface, that is, between the recording time of the channel knowledge map and the user's collection time, the first... The change in the reflective surface corresponding to the historical mirror base station type is the first The reflective surface corresponding to each real-time mirror base station class, and the changes can be based on the first... In the historical mirror base station class The distribution of the estimated positions of the nth reflection point and the nth reflection point In each real-time mirror base station class The distribution of estimated locations of individual reflection points is used to infer and calculate. .

[0109] The application discloses a wireless communication environment change perception method based on a channel knowledge map, uses a single-path optimization algorithm to extract path information of each reflection path from channel information of each user and calculate an actual departure direction and an actual arrival direction, constructs a maximum likelihood estimation problem with the aim of minimizing noise error, introduces a penalty function to process coplanar constraints and construct a multi-convex problem, adopts alternating iterative optimization to obtain an optimal departure direction and an optimal arrival direction of each reflection path of each user, calculates a corresponding reflection point initial estimation position with reference to a base station position and a position of each user, and effectively reduces dimensionality of high-dimensional channel information of multiple paths of each user through the reflection point initial estimation position; based on a mirror consistency principle, a clustering correction algorithm is adopted to calculate real-time mirror base station positions of each reflection path of each user in each iteration, determine all real-time mirror base station classes and corresponding real-time centroid positions through clustering, correct the reflection point initial estimation position of each reflection path in each real-time mirror base station class based on the real-time centroid position of each real-time mirror base station class, repeat iteration until convergence, obtain reflection point estimation positions of each reflection path in each real-time mirror base station class, and weaken the influence of collected channel information error on the reflection point position estimation; all historical mirror base station classes and reflection point estimation positions of each reflection path in each historical mirror base station class are obtained through processing of the map information set by the single-path optimization algorithm and the clustering correction algorithm, all real-time mirror base station classes are matched with all historical mirror base station classes based on a minimum matching principle, and reflection surface change perception in a high-precision wireless communication environment is realized.

[0110] The above is only the preferred embodiment of the present application, and the protection scope of the present application is not limited to the above-mentioned embodiment. Any technical solution falling within the concept of the present application shall be deemed to fall within the protection scope of the present application. It should be noted that, for ordinary skilled persons in the art, some improvements and refinements without departing from the principle of the present application shall also be deemed to fall within the protection scope of the present application.

Claims

1. A method for wireless communication environment change perception based on channel knowledge map, characterized in that, The method comprises the following steps: From base stations in local areas During the communication process of each user, location and channel information are obtained to construct a user information set, and local area information is retrieved from the channel knowledge map. The location and channel information of each grid point are used to construct a map information set; Using a single-path optimization algorithm, starting from the first path... Extracting the channel information of the user from the first user The path information of each reflection path is obtained, and the actual departure and arrival directions are calculated. A maximum likelihood estimation problem is constructed with the objective of minimizing noise error. A penalty function is introduced to handle coplanar constraints, constructing a multiconvex problem. Alternating iterative optimization is used to obtain the solution for the th reflection path. The first user's The optimal departure and arrival directions of the reflection path are referenced to the base station location and the first... The location calculation of the first user The first user's The initial estimated location of the reflection point along the reflection path. , This represents the number of reflection paths. Based on the mirror consistency principle, a clustering correction algorithm is adopted to calculate the real-time mirror base station position of each user for each reflection path in each iteration, to determine the real-time mirror base station class and the corresponding real-time centroid position through clustering, to correct the reflection point initial estimation position of each reflection path in the real-time mirror base station class based on the real-time centroid position of the first real-time mirror base station class, and to repeat the iteration until convergence, so as to obtain the reflection point estimation position of each reflection path in the first real-time mirror base station class, is the total number of reflection surfaces which is the same as the number of reflection paths.​​​​​ The map information set is processed by a single-path optimization algorithm and a clustering correction algorithm to obtain a historical mirror base station class and a reflection point estimated position of each reflection path in each historical mirror base station class, and each real-time mirror base station class is matched with each historical mirror base station class based on a minimum matching principle, and a change of a reflecting surface in a wireless communication environment is tested. The single-path optimization algorithm extracts path information of the first reflected path from channel information of the first user, including path gain, elevation arrival angle, horizontal arrival angle, elevation departure angle and horizontal departure angle of the first reflected path of the first user, calculates actual departure direction and actual arrival direction of the first reflected path of the first user based on stereographic geometry, and satisfies coplanar constraint and modulus constraint for accurate departure direction and accurate arrival direction of the first reflected path of the first user, wherein the coplanar constraint requires that the accurate departure direction, the accurate arrival direction of the first reflected path of the first user are coplanar with the line connecting the positions of the first user and the base station, and the modulus constraint requires that the modulus of the accurate departure direction, the accurate arrival direction of the first reflected path of the first user is 1, so as to minimize the error between the departure direction, the arrival direction of the first reflected path of the first user and the actual departure direction, the actual arrival direction of the first reflected path of the first user, and the maximum likelihood estimation problem is constructed in combination with the coplanar constraint and the modulus constraint. Based on the minimum matching principle, calculate the... The estimated position of the reflection point for each reflection path in each real-time mirror base station class is compared with the first... The Hasdorf distances to the estimated reflection points of each reflection path in each historical mirror base station class are used to construct a Hasdorf distance matrix. The first row of the Hasdorf distance matrix is ​​then... Each column is treated as a first-level node. Following the principle of no repeated columns, each row of the Hasdorf distance matrix is ​​explored sequentially. During the exploration of the [number]th [row]... When the line is drawn, for the first line... Each row Level node, excluding rows 1 to 1 Rows of selected columns, based on unselected columns Each column is for Level node generation indivual Level nodes, until the exploration reaches the first level. Okay, we've found the distance to the forest. Select the optimal path that minimizes the total distance to Hasdorf in the forest, and determine the optimal path based on the number of columns corresponding to each level node in the optimal path. Real-time mirror base station class and The matching relationship of each historical mirror base station class is determined by querying the Hasdorf distance matrix to obtain the corresponding Hasdorf distance, which is then compared with the distance threshold to decide whether the reflective surface corresponding to each matched historical mirror base station class and the real-time mirror base station class has changed.

2. The channel knowledge map based wireless communication environment change awareness method of claim 1, wherein, In coplanar constraints, the first The first user's The direction of departure of the reflection path and direction of arrival The coupling causes the maximum likelihood estimation problem to be nonconvex, which can be addressed by using a penalty function. By handling the coplanar constraints and adding them to the optimization objective of the maximum likelihood estimation problem, a multiconvex problem is constructed.

3. The channel knowledge map based wireless communication environment change awareness method of claim 2, wherein, The alternating iterative optimization method is used to obtain the first... The first user's The optimal departure and arrival directions of the reflection path are determined by the following steps: In the In the round of iteration, the fixed number of rounds is... The arrival direction of the wheel is substituted into the multiconvex problem, simplifying it to generate the first subproblem. The modulus constraint is simplified to a modulus of 1 in the departure direction. The first Euclidean gradient of the first convex function in the first subproblem is solved using the Riemannian manifold method, and then projected onto the tangent space of the unit modulus manifold to obtain the first... The first Riemann gradient of the cycle, according to the first Wheel departure optimization step size correction The direction of the wheel's departure, reprojected to obtain the first... The direction of wheel departure is determined, and the first wheel is determined based on the line search criterion and step size limitation ratio. The wheel's departure step size is optimized; Fixed number Substituting the departure direction of the wheel into the multiconvex problem simplifies to generate a second subproblem, and the module constraint is simplified to a module length of 1 in the arrival direction. The Riemannian manifold method is then used to correct the second subproblem. The direction of arrival of the wheel is reprojected to obtain the first... The direction of arrival of the wheel and the determination of the first Optimize the arrival step size of the wheel; The direction of departure and the direction of arrival of the wheel are substituted into the multi-convex function to obtain a multi-convex function value of the wheel The direction of departure and the direction of arrival of the wheel are substituted into the multi-convex function to obtain a multi-convex function value of the wheel The direction of departure and the direction of arrival of the wheel are substituted into the multi-convex function to obtain a multi-convex function value of the wheel The direction of departure and the direction of arrival of the wheel are substituted into the multi-convex function to obtain a multi-conv If the value is less than the first convergence threshold, stop the iteration and set the... The direction of departure and the direction of arrival of the wheel are used as the first The first user's If the optimal departure and arrival directions of the reflection path are greater than or equal to the first convergence threshold, determine the current round number. Is it equal to the maximum number of rounds? If equal to the maximum number of iterations, stop the iterations and set the direction of departure and the direction of arrival of the first hop as the optimal direction of departure and the optimal direction of arrival of the first reflection path for the first user, if less than the maximum number of iterations, start the first iteration.

4. The channel knowledge map based wireless communication environment change awareness method of claim 3, wherein, determining a departure optimization step and an arrival optimization step for the vehicle, comprising the steps of: wheel, comprising the steps of: The first Substituting the departure and arrival directions of the wheel into the first and second convex functions respectively, we obtain the... The first and second convex function values ​​of the wheel; The first The first convex function value of the wheel plus the step size shrinkage ratio, the second... The departure optimization step size and the first wheel The product of the first Riemann gradient of the rounds yields the product of the second Riemann gradient. The first convex threshold of the wheel; add the product of the step-size reduction ratio, the first round second convex function value, the first round arrival optimization step, and the first round second Riemann gradient to obtain the first round second convex threshold value; According to the line search criterion, based on the first convex function value of the first wheel, whether the first convex threshold value of the first wheel is greater than the first convex threshold value of the first wheel, the first convex threshold value of the first wheel is selected, and the leaving optimization step of the first wheel is set as the product of the step reduction ratio and the leaving optimization step of the first wheel or the leaving optimization step of the first wheel is continued. based on whether the second convex function value of the wheel is greater than the first convex threshold value of the wheel selecting the first wheel to the product of the step size reduction ratio and the arrival optimization step size of the first wheel or continuing the arrival optimization step size of the first wheel.

5. The channel knowledge map based wireless communication environment change awareness method of claim 1, wherein, Based on the principle of mirror consistency, the results are calculated using the precise arrival and departure directions. The user in the first The real-time mirror base station locations of the reflection paths of each reflector coincide, calculated based on the optimal departure and arrival directions. The user in the first The real-time mirror base station locations of the reflection paths of the first reflecting surface are distributed within a local neighborhood, constituting the first... The real-time mirror base station class, for the first The first user's The reflection path, calculate the first one. The sum of the reflection distance between the location of the first user and the initially estimated location of the reflection point, and the incident distance between the location of the base station and the initially estimated location of the reflection point, yields the first... The first user's The propagation distance of the reflection path, the first The first user's The real-time mirror base station location of the reflection path is equal to the first The product of the location of each user, the optimal direction of arrival, and the propagation distance is calculated. The real-time mirror base station locations for each reflection path of each user are used to construct a real-time mirror base station location set.

6. The channel knowledge map based wireless communication environment change awareness method of claim 5, wherein, The cluster correction algorithm comprises the following steps: The cluster correction algorithm comprises the following steps: The real-time mirror base station position set is taken as the clustering set of the 0th round, and the optimal arrival direction and the optimal departure direction of each reflection path of the user are taken as the direction set of the 0th round. The real-time mirror base station position set is taken as the clustering set of the 0th round, and the optimal arrival direction and the optimal departure direction of each reflection path of the user are taken as the direction set of the 0 In the In the round of iteration, a clustering algorithm is used to classify the th... Clustering the cluster sets of the rounds yields the first... Wheel The first real-time mirror base station class and the first The set of the center of mass positions of the wheel; Based on the first Wheel of the first Real-time mirror base station class centroid position and each user's location to calculate each user in the first Reflection surface of the first Wheel of the initial direction of arrival, according to the first Wheel of the first Real-time mirror base station class centroid position and base station location to calculate the first Reflection surface of the first Wheel normal vector, based on the geometric relationship to calculate each user in the first Reflection surface of the first Wheel of the initial direction of departure, get the first Wheel of the initial direction set, ; Based on the update ratio The first The initial direction set of the wheel and the first The weighted summation of the direction sets of the wheels yields the first... The direction set of the wheel, calculation In each of the real-time mirror base station classes, each reflection path is in the... Estimate the position of the reflection point of the wheel and deduce the corresponding first... The cluster set of wheels; computing a cluster set of the first wheel and a cluster set of the second wheel and determining whether an average distance error of all real-time mirror base station positions in the cluster set of the second wheel is less than a second convergence threshold; If less than the second convergence threshold, stopping iteration and output The estimated position of the reflection point of each reflection path in the real-time mirror base station class in the first round, if greater than or equal to the second convergence threshold, whether the current round number is equal to the maximum clustering round number; If equal to the maximum clustering round, stop iteration and output The estimated position of the reflection point of each reflection path in the real-time mirror base station class in the first round, if less than the maximum clustering round, start the first round iteration.

7. The channel knowledge map based wireless communication environment change awareness method of claim 3, wherein, The optimal departure direction and the optimal arrival direction of the first reflection path of the first user are obtained by alternately iteratively optimizing The optimal departure direction and the optimal arrival direction of the first reflection path of the first user are obtained by alternately iteratively optimizing The optimal departure direction and the optimal arrival direction of the first reflection path of the first user are obtained by alternately iteratively optimizing The optimal departure direction and the optimal arrival direction of the first reflection path of the first user are obtained by alternately iteratively optimizing The optimal departure direction and the optimal arrival direction of the first reflection path of the first user are obtained by alternately iteratively optimizing The optimal departure direction and the optimal arrival direction of the first reflection path of the first user are obtained by alternately iteratively optimizing

Images