A single-transmitter and single-receiver three-dimensional visible light positioning method based on rotation

Abstract

The present invention provides a single-transmit and single-receive three-dimensional visible light positioning method based on rotation, comprising: step 1, positioning scene parameterization step; using x min with x max Indicates the minimum and maximum coordinates that the UE can reach in the horizontal direction in the scene, y min with y max Indicates the minimum and maximum coordinates that the UE can reach in the vertical direction in the scene, z min With z max Indicates the minimum and maximum coordinates that the UE can reach in the height direction in the scene; Step 2, the rotation positioning step; the positioning process is performed using the channel response obtained in different postures during the rotation process; Step 3, the step of solving the non-convexity problem; the particle swarm optimization algorithm is introduced to solve the positioning error problem caused by the non-convexity of the positioning process in Step 2, and the final positioning result is obtained. The beneficial effects of the present invention are: 1. The method of the present invention can be used in ISAC indoor positioning scenarios; 2. The method of the present invention allows a UE equipped with a single PD to rotate multiple times to complete the positioning process under a single light condition.

Description

Technical Field

[0001] The present invention relates to the field of communication positioning technology, and in particular to a single-transmitting and single-receiving three-dimensional visible light positioning method based on rotation. Background Art

[0002] With the advent of the Industrial Internet of Things (IIoT), more and more devices are demanding high-precision indoor positioning, leading to widespread attention for various indoor positioning technologies. Visible light positioning is a key component of high-precision indoor positioning. Compared to common indoor positioning technologies such as Wi-Fi (Wireless Fidelity), Bluetooth, and ultra-wideband, visible light positioning offers advantages such as low power consumption, low deployment costs, and high positioning accuracy. However, existing visible light positioning methods require a large number of receivers or transmitters, making them difficult to implement in practical scenarios where the number of transmitters and receivers is insufficient.

[0003] Existing visible light technologies utilize light-emitting diodes (LEDs) as transmitters and multiple photodiodes (PDs) or complementary metal oxide semiconductors (CMOS) at the receiver end to receive the LED signals. Positioning is performed based on signal characteristics such as signal strength, reception time, and reception angle. Existing visible light technologies rely on the availability of multiple LEDs, PDs, or CMOS. However, most existing terminal devices only have a single PD, which makes positioning impossible when indoor lighting resources are insufficient.

[0004] References:

[0005]

[55] ZHOU B, LIU A, LAU V.Performance Limits of Visible Light-based UserPosition and Orientation Estimation Using Received Signal Strength Under NLOSPropagation[J]. IEEE Transactions on Wireless Communications, 2019, 18(11): 5227-5241.

[0006]

[56] YU

[0007]

[57] YU Summary of the Invention

[0008] The present invention provides a rotation-based single-transmit and single-receive three-dimensional visible light positioning method, comprising:

[0009] Step 1, positioning scene parameterization step; use x min with x max Indicates the minimum and maximum coordinates that the UE can reach in the horizontal direction in the scene, y min with y max Indicates the minimum and maximum coordinates that the UE can reach in the vertical direction in the scene, z min With z max Indicates the minimum and maximum coordinates that the UE can reach in the height direction of the scene.

[0010] Step 2, rotation positioning step; use the channel responses under different postures obtained during the rotation process to perform the positioning process, that is, find a point in the spatial position so that the Euclidean distance between its theoretical channel response value under K postures and the channel response value actually obtained through channel estimation during the rotation process is minimized.

[0011] Step 3: Solve the non-convexity problem; introduce the particle swarm optimization algorithm to solve the positioning error problem caused by the non-convexity in the positioning process of step 2 and obtain the final positioning result.

[0012] As a further improvement of the present invention, in step 2, the positioning process can be regarded as an optimization problem, which can be expressed as:

[0013]

[0014] In the formula ——Final positioning result of the R-VLP system;

[0015] ——Any point in the solution space;

[0016] ——The theoretical channel response value at any point in the kth posture;

[0017] h k ——The channel response value under the k-th posture estimated by the channel during the actual rotation process;

[0018] The constraint C1 in formula (3-17) indicates that the possible positioning results should exist within the coverage range of the LED light, that is, the final positioning search range is a cone range from the LED downward. The constraints C2-C5 indicate that the search should be within the set fixed range, and then through the limited range x min with x max 、y min with y max and z min With z max To search.

[0019] As a further improvement of the present invention, the specific steps of step 3 include:

[0020] Step 30: Input step: Input the channel estimation results under K rotation postures (k=0,1,…,K), particle swarm size N and number of particle swarm iterations M.

[0021] Step 31, initialization step: In the initial stage of the particle swarm algorithm, it is necessary to initialize the incident vector r corresponding to N particles n，0 , and give each particle an initial velocity vector v n，0 To complete the subsequent iterative process, the subsequent process is to iterate the entire particle swarm M times.

[0022] Step 32, the positioning result output step; the incident vector corresponding to the historical optimal particle of the group represents the historical optimal position of the group Output as the final positioning result of R-VLP.

[0023] As a further improvement of the present invention, step 31 specifically includes:

[0024] Step 1: Calculate the current fitness value ζ of each particle n .

[0025] Step 2: Based on the latest fitness value ζ n , update the optimal position information of the group and update the particle movement speed vector v n，m .

[0026] Step 3, through the velocity vector v n，m , get the incident vector r corresponding to each particle in the next iteration n,m , so far one iteration is completed, and then return to step 1 and repeat.

[0027] As a further improvement of the present invention, in step 1, the objective function of the optimization problem (3-17) can be used to calculate the fitness value of the nth particle, and the calculation formula is as follows:

[0028]

[0029] The fitness value represents the possibility that the position corresponding to the current particle is the final positioning result. In the subsequent process, the movement of the particle can be guided according to the fitness value.

[0030] As a further improvement of the present invention, the step 2 includes:

[0031] In the first step, the latest optimal historical position of each particle is selected based on its current fitness value and its historical fitness value in the previous iteration. The process is as follows:

[0032]

[0033] in It represents the number of iterations corresponding to the historical optimal position of the nth particle, and the corresponding historical optimal incident vector is

[0034] In the second step, the historical optimal vector of the entire particle swarm is selected through the historical optimal incident vector of each particle. The process is as follows:

[0035]

[0036] in It represents the number of the particle that has reached the historical optimal position of the group in the entire particle swarm, and its corresponding historical optimal incident vector is

[0037] Step 3: Calculate the moving velocity vector v of the nth particle in the current iteration n,m , the calculation formula is as follows:

[0038]

[0039] Among them, c1 and c2 are two non-negative constants, which represent the proportion of the particle's own historical optimal position information and the group's historical optimal position information in the position update process. r1 and r2 represent two random variables that obey a uniform distribution in the closed space from 0 to 1.

[0040] As a further improvement of the present invention, the step three is specifically as follows:

[0041] The incident vector r n，m Update it to r according to the following formula n，m+1

[0042] r n，m =r n,m-1 +v n,m-1 (3-31).

[0043] As a further improvement of the present invention, in step 32, a fixed number of particle swarm iterations M is specified in the R-VLP. After M iterations, the final positioning result is obtained by the incident vector corresponding to the historical optimal position information of the swarm. The formula is as follows:

[0044]

[0045] The beneficial effects of the present invention are: 1. The method of the present invention can be used in indoor positioning scenarios of ISAC; 2. The method of the present invention allows a UE equipped with a single PD to rotate multiple times in a single-lamp situation to complete the positioning process; 3. The method of the present invention is robust to the number of hardware devices and can complete the work in any indoor scenario. In this respect, it is far superior to existing methods that require at least three LEDs or three PDs, such as: trilateration method, fingerprint method and approximation method. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 Schematic diagram illustrating parameters in the visible light positioning channel model of the present invention;

[0047] Figure 2 It is a schematic diagram illustrating the rotation model of the present invention and its parameters;

[0048] Figure 3 Schematic diagram of the relationship between UE normal vectors in adjacent postures of the present invention;

[0049] Figure 4 It is a graph proving the non-convexity of the optimization problem in the R-VLP system of the present invention;

[0050] Figure 5 This is a schematic diagram of a user performing static positioning in a corridor environment according to the present invention;

[0051] Figure 6 This is a performance comparison chart of the R-VLP of the present invention with other single-lamp positioning systems and traditional RSS systems;

[0052] Figure 7 This is a graph showing the effect of particle swarm size on the positioning accuracy of the R-VLP system of the present invention;

[0053] Figure 8 This is a graph showing the effect of the number of particle group movements on the positioning accuracy of the R-VLP system of the present invention;

[0054] Figure 9 This is a graph showing the effect of the number of particle group movements on the positioning accuracy of the R-VLP system of the present invention;

[0055] Figure 10 1 is a diagram showing the approximation of the CRLB by R-VLP under different parameter combinations and random rotations of the present invention;

[0056] Figure 11 It is a graph showing the effect of the number of postures on the R-VLP positioning accuracy in the rotation model of the present invention;

[0057] Figure 12 is a diagram showing the effect of the incident angle on the R-VLP positioning accuracy in the rotation model of the present invention;

[0058] Figure 13 is a diagram showing the effect of the rotation angle on the R-VLP positioning accuracy in the rotation model of the present invention;

[0059] Figure 14 This is a flow chart of the particle swarm-based R-VLP algorithm of the present invention. DETAILED DESCRIPTION

[0060] Glossary:

[0061] VLP: visible light positioning; R-VLP: rotational visible light positioning; PD: photodiode

[0062] To address the need for single-transmitter, single-receiver static positioning, the present invention proposes a rotation-based single-transmitter, single-receiver three-dimensional visible light positioning method. Based on the received signal strength, a visible light positioning model for terminal rotation is established, and an optimization problem for visible light positioning is constructed using this model. Due to the non-convexity of this optimization problem, a particle swarm optimization algorithm is introduced to achieve centimeter-level positioning accuracy with lower complexity. Combining the constructed optimization problem, a closed-form solution to the Cramer-Rao lower bound of its positioning error is derived, and the influence of the rotation angle on positioning accuracy is analyzed. Simulation results show that the accuracy of this positioning method is better than that of other existing single-LED scenarios under the same environmental parameter settings. With the reduction of the distance between the transmitter and receiver, the increase in the size and number of cycles of the particle swarm, the increase in the number of rotation postures, and the expansion of the rotation angle and incident angle, this rotation positioning method can achieve higher positioning accuracy.

[0063] The present invention is applied to visible light indoor positioning scenarios. When the terminal has the ability to rotate, positioning can be performed through a single PD at the receiving end and a single LED in the scene.

[0064] The present invention first describes in Section 3.1 the positioning model in the rotation positioning method proposed to solve the problem of indoor VLP system's dependence on hardware resources. In this model, the rotation process is abstracted into a behavior that is only related to two parameters, the rotation angle and the deflection angle. This facilitates the qualitative description of the rotation process through subsequent simulation experiments. Section 3.2 of the present invention describes the VLP combined with the rotation model, that is, the rotating visible light positioning system model, which mainly includes the transformation of the Rembrandt channel model, the modeling of the positioning system and the derivation of the CRLB of the model. Section 3.3 of the present invention describes the particle swarm algorithm, which mainly includes the reasons for selecting the particle swarm algorithm to solve the non-convex problem of the rotation visible light positioning (Rotation-VLP, R-VLP) system and the principle of the particle swarm algorithm. Section 3.4 simulates and verifies the positioning performance of the model proposed by the present invention after applying the particle swarm algorithm, as well as the performance under different system parameters.

[0065] 3.1 Receiver Rotation Model

[0066] Different from the time-varying channel of radio frequency, the visible light channel is a deterministic time-invariant system model that depends on many parameters, such as irradiation angle, incident angle, and the distance between LED and PD, e.g. Figure 1 As shown. This means that when the visible light channel between the LED and the PD remains unchanged, changing any parameter in the channel model will cause the visible light channel to change accordingly. In traditional RSS positioning, different received signals are used for positioning, and the traditional VLP method cannot locate when there is insufficient lighting resources due to the insufficient number of different signals. Therefore, the characteristic that this time-invariant channel only follows parameter changes can be used to artificially create different received signal situations when there is insufficient light source or insufficient number of PDs. This idea prompted the present invention to propose a novel R-VLP system. The method of the present invention can be used in the indoor positioning scenario of ISAC. The method allows a UE equipped with a single PD to rotate multiple times in a single-lamp situation to complete the positioning process. The method is robust to the number of hardware devices and can complete the work in any indoor scenario. In this respect, it is far superior to existing methods that require at least three LEDs or three PDs, such as triangulation methods, fingerprint methods, and approximation methods.

[0067] The present invention considers an indoor three-dimensional VLP system, in which there is only one UE equipped with a single PD as a receiving device and an LED on the ceiling as a transmitting device. UE =[x UE ,y UE , z UE ] represents the UE's position vector, where xUE ,y UE and z UE Represent the horizontal, vertical and height coordinates of UE respectively, Represents the normal vector of UE, that is, the vertical normal vector of PD plane, where and Respectively represent the horizontal, vertical and height coordinates of the normal vector, with p LED =[x LED ,y LED , z LED ] represents the position vector of the LED, where x LED ,y LED and z LED Represent the horizontal, vertical and height coordinates of the LED respectively, Represents the normal vector of the LED, that is, the vertical normal vector of the LED plane, where and Represents the horizontal, vertical and height coordinates of the normal vector respectively.

[0068] Based on the above basic system model, a rotation method is proposed to provide more channel information when hardware resources are insufficient. The rotation process can be achieved by Figure 2 To describe, assume that the position of UE remains unchanged during the rotation, and introduce a new incident vector r to represent the vector from LED to UE, which is expressed as follows

[0069] r=p UE -p LED (3-1)

[0070] Where r is the incident vector between the transmitting LED and the receiving PD, r = [x r ,y r , z r ], x r ,y r and z r Represents the horizontal, vertical, and altitude coordinates of the incident vector.

[0071] The distance d between the LED and the UE can be expressed as

[0072]

[0073] The entire process is that the UE first deflects from a state where the incident vector is perpendicular to the incident vector, and the deflection angle is Ψ. This deflection obtains the first posture in the rotation process. Then, it rotates around the axis between the LED and the UE by a certain angle, which is Υ. This rotation process obtains the second posture in the rotation process. Assuming that this rotation process is performed K-1 times in total, K different postures will be obtained. The deflection angle and rotation angle of each posture are not required to be fixed during the rotation process. That is, the k-th posture can have independent deflection angles Ψ. k And independent rotation angle Υ k，k+1 However, in order to illustrate the influence of parameters on positioning effect in subsequent simulations, the same deflection angle and rotation angle are used for K different postures in a rotation process in the simulation experiment. Under this premise, except for the change of the UE normal vector during the rotation process, other angle parameters remain fixed. The UE normal vector in the kth posture is expressed as q UE，k , so that the end points of all K UE normal vectors during the entire rotation process form a circle around the incident vector r.

[0074] according to Figure 1 According to the geometric law in the irradiation angle θ, it can be expressed as

[0075]

[0076] The incident angle Ψ in the kth posture during the rotation process k The incident angle at the UE end can be expressed as

[0077]

[0078] Assume that all normal vectors are unit vectors (their length does not affect the subsequent formula derivation), that is, ||q UE，k ||=1 and ||q LED ||=1, after substituting formula (3-1) and formula (3-2) into formula (3-3) and formula (3-4), the irradiation angle θ and the incident angle Ψ in the kth posture are k It can be expressed as follows

[0079]

[0080] As mentioned in the above description, each rotation posture can have its own independent rotation angle Y k，k+1 , which describes the rotation angle between the kth posture and the k+1th posture. In order to further describe the UE normal vector q under two adjacent postures UE,k With q UE,k+1 The relationship between the UE normal vector q under the k-th posture UE,k according to Figure 3 After decomposition, it can be expressed as:

[0081] q UE,k =q par +q ver,k (3-7)

[0082] Where q par — parallel normal vector,

[0083] q ver,k ——The vertical normal vector at the k-th pose.

[0084] Now by introducing formula (3-6), we can obtain the parallel normal vector q par The new expression is as follows

[0085]

[0086] The vertical normal vector q in the kth posture ver,k The expression of can be obtained by substituting formula (3-8) into formula (3-7), and its expression is as follows

[0087]

[0088] Because during the rotation, the incident angle Ψ at each posture is assumed to be k are equal, and the rotation angle Ψ between any two postures k，k+1 is also equal, in this case when rotating, the parallel normal vector q par is constant, so we only need to obtain the vertical normal vector q under two adjacent postures ver，k With q ver，k+1 The relationship between the two adjacent postures can be expressed as the UE normal vector q UE，k With q UE，k+1 relationship, such as Figure 3 As shown, define a normal vector q perpendicular to the k-th posture ver，k The perpendicular vector to the incident vector r It can be expressed as:

[0089]

[0090] In the formula ——Cross product operation between vectors. Figure 3 It can be seen that the vertical normal vector q in the k+1th posture ver，k+1 is the vertical normal vector q under the k-th posture ver，k Rotate Y around the incident vector r k，k+1 Therefore, q ver，k+1It can be expressed as q ver，k The corresponding perpendicular vector The linear combination of , which can be expressed as:

[0091]

[0092] At this time, substituting formula (3-9) and formula (3-10) into formula (3-11), we can obtain the following relationship:

[0093]

[0094] Because q UE，k+1 In essence, it can be expressed as its parallel normal vector q par Its perpendicular normal vector q ver，k+1 The linear combination of , so we can get the following expression:

[0095]

[0096] For the incident vector r in formula (3-10) and the UE normal vector q in the kth posture UE，k The cross product operation between them can be rewritten as the following vector representation:

[0097]

[0098] Substituting formula (3-14) into formula (3-13), we can get the final q UE，k With q UE，k+1 The expression between them is:

[0099]

[0100] 3.2 Rotating visible light positioning system model

[0101] Section 3.1 introduced the rotation model. This section introduces the indoor R-VLP model combined with the rotation model, including the transformation of the Rembrandt channel, the description of the positioning model, the description of its non-convexity, and the derivation of the Cramer-Rao lower bound for this system.

[0102] 3.2.1 Adjusted Rembrandt channel model and positioning model

[0103] In the Rembrandt channel model formula (2-3) of the visible light channel, the UE location information p UE It is hidden in the distance information d, the incident angle information cos(Ψ) and the irradiation angle information cos(θ). In the rotation model of the present invention, only the UE normal vector q occurs during the positioning process. UE，kThe original Rembrandt channel model cannot express this change because the UE normal vector information is also hidden in the angle information cos(Ψ) and cos(θ). Therefore, the Rembrandt channel model needs to be modified to enable it to reflect the rotation characteristics. Finally, the modified Rembrandt channel model can be used to obtain K unrelated channel formulas through the rotation process.

[0104] In the above context, after substituting formula (3-2), formula (3-5) and formula (3-6) into formula (2-3), assuming the Rembrandt order of the LED is m Lam = 1 (this assumption is based on the fact that most LED half-power angles are approximately 60°), the following new Rembrandt channel model can be obtained:

[0105]

[0106] Where L is the visible light channel constant between the transmitting LED and the receiving PD, L = -A rd / π;

[0107] Q k ——The pose correlation matrix between the transmitting LED and the receiving PD,

[0108] Since the position of the UE does not change during the rotation, the incident vector r does not change during the rotation. Therefore, the visible light channel in each rotation state is mainly affected by the normal vector cross-correlation matrix Q k In order to carry out the positioning process of the UE, the channel impulse response under the current posture needs to be estimated by MMSE in each rotation process. As shown in formula (3-16), each rotation will generate an independent channel, so the specific postures in the K rotation processes correspond to K independent channels, which can be used for positioning. Compared with the visible light positioning method based on multiple PDs, this rotation positioning method can reduce the requirements for the receiving device on the terminal device, that is, no additional PD is required on the terminal. In addition, the R-VLP system proposed in the present invention is also applicable to the case of a single LED, relaxing the strict requirements on the number of LEDs in the traditional VLP system.

[0109] In order to propose a positioning model problem, it is necessary to parameterize the positioning scene first. First, use x min with x max Indicates the minimum and maximum coordinates that the UE can reach in the horizontal direction in the scene, y min with y max Indicates the minimum and maximum coordinates that the UE can reach in the vertical direction in the scene, z min With z maxIndicates the minimum and maximum coordinates that the UE can reach in the height direction of the scene. Next, the channel response obtained in different postures during the rotation process can be used for positioning. The positioning process can be regarded as an optimization problem, that is, to find a point in space so that the Euclidean distance between its theoretical channel response value in K postures and the channel response value actually obtained through channel estimation during the rotation process is the smallest. This optimization problem can be expressed as

[0110]

[0111] In the formula ——Final positioning result of the R-VLP system;

[0112] ——Any point in the solution space;

[0113] ——The theoretical channel response value at any point in the kth posture.

[0114] h k ——The channel response value under the k-th posture estimated by the channel during the actual rotation process.

[0115] The constraint C1 in the optimization problem indicates that the possible positioning results should exist within the coverage of the LED light, that is, the final positioning search range is a cone range from the LED downward, rather than a complete cube. The constraints C2-C5 indicate that the search should be within a set fixed range. Because in actual scenes, the coverage of a single light is always limited, so it is impossible to search for the final positioning result from an infinite space. Instead, the final positioning result is obtained based on the LED position information p. LED To complete a rough position estimate, and then by limiting the range x min with x max 、y min with y max and z min With z max To search.

[0116] 3.2.2 Description of the non-convexity of the positioning model

[0117] In Section 3.2.1, an optimization problem for the positioning process is proposed. However, the problem is that the optimization problem is non-convex, which means that there are several local optimal values in the solution space. Traditional linear optimization algorithms, such as the Gaussian iteration method, are prone to falling into the local optimal solution in this problem, thus obtaining an incorrect positioning result. Its non-convexity can be proved theoretically, but in order to intuitively express this non-convex characteristic, the process of calculating the Euclidean distance in formula (3-17) is performed on several points in a two-dimensional plane, and the inverse of the Euclidean distance calculated for each point is displayed in Figure 4 In the figure, we can clearly see that there are two local optimal solutions on only one two-dimensional plane, and the final desired positioning result is actually the global optimal solution on the plane.

[0118] In order to solve this optimization problem, a particle swarm algorithm for solving non-convex optimization problems will be introduced in Section 3.3 of this invention. However, there are actually some other non-convex optimization algorithms, such as genetic algorithms and annealing algorithms. The reason for choosing the particle swarm algorithm will also be explained in Section 3.3.

[0119] In order to estimate the positioning performance of the R-VLP of the present invention before applying a specific algorithm and derive the CRLB of the positioning results under this model, the results after using the particle swarm solver algorithm will be compared with the CRLB, and the effect of parameter approximation will be explained.

[0120] In the derivation of the formula of the present invention, a Gaussian model is used to describe the UE position p UE The prior distribution of where χ prior Represents the precision matrix, also p UE The inverse matrix of the autocorrelation matrix, and the normal q of UE UE The vector hypothesis can be obtained from the IMU.

[0121] In the actual process, we will have prior knowledge about the UE location parameters from the LED. In other words, when we know the position of the LED that is communicating with the UE, since the lighting range of the LED is limited, we can use this feature to roughly estimate the UE's location range.

[0122] Proposition 3-1: If there are K poses obtained from the rotation process in the R-VLP system, then the final positioning result p obtained by the R-VLP system is UE The mean square error of has the following form of CRLB bound

[0123]

[0124] Where E{·} is the mathematical expectation;

[0125] trace——indicates the trace of the matrix;

[0126] ——CRLB of the positioning accuracy of the R-VLP system, its expression is as follows.

[0127]

[0128] Where A k ——The rotation information matrix, its expression will be given in the subsequent proof process.

[0129] prove:

[0130] According to the signal estimation and detection theory

[59] , the position vector p of the UE in the R-VLP system UE There is a theoretical limit to the unbiased estimate of The derivation process is as follows. First, we give The following general form

[0131]

[0132] In the formula ——position prior information,

[0133] ——Rotation information, its specific expression is as follows

[59] .

[0134]

[0135] In the formula ——Indicates that p UE Find the second-order partial derivative;

[0136] ——Represents the channel matrix vector obtained after channel estimation

[0137] ——The likelihood function corresponding to the PD channel estimation value at the receiving end is expressed as follows

[0138]

[0139] In the formula, H——at p UE The theoretical channel response vector under K postures on the potential solution of

[0140] ——Estimated channel matrix The difference matrix with the theoretical channel matrix h,

[0141] C——channel matrix estimated by channel and The covariance matrix between .

[0142] At this point, you can give The closed-form solution form of the rotation information matrix A k Given by the following formula

[0143]

[0144] In the formula ——Horizontal position information brought by the rotation process;

[0145] ——Vertical position information brought by the rotation process;

[0146] ——Altitude position information brought by the rotation process;

[0147] ——Horizontal and vertical position information brought by the rotation process;

[0148] ——Horizontal and height-related position information brought by the rotation process;

[0149] ——Vertical and height-related position information brought about by the rotation process.

[0150] Rotation information matrix A k It mainly depends on the choice of posture during the rotation process, which is essentially the theoretical channel response h under the k-th posture k The rotation information matrix represents the channel matrix vector estimated by the existing channel, and the system has the following equations for each potential UE position information p: UE A zero-resolution rotation information matrix means unpredictability of parameters, because for different potential UE position information p UE For example, the theoretical channel response value h k are the same, which means that it is impossible to use this parameter to achieve the desired positioning effect. On the contrary, the system's ability to distinguish potential solutions can be enhanced by reducing the correlation between poses or increasing the number of poses. Related enhancement methods will be discussed later.

[0151] If the independent rotation information matrix A in formula (3-19) is k Expressed as a related form, we can substitute into formula (3-15) to get all A k Represented as the initial posture q UE，1With the rotation angle γ k，k+1 The related form, and the two adjacent UE normal vectors q can be obtained by formula (3-15) UE，k With q UE，k+1 The correlation between

[0152]

[0153] The last part of the above formula can be proved to be zero. The proof process is as follows

[0154]

[0155] At the same time, the normal vectors are unit vectors that only represent directions, so q UE，k With q UE，k+1 The correlation between can be rearranged as

[0156]

[0157] When only K=3 is selected during the rotation process, that is, when only three postures are selected, the rotation angle Υ k，k+1 The range of change is (0°, 120°], at this time cos(Υ k，k+1 ) changes from 1 to -0.5. Due to the monotonicity of the function, it is known that q UE，k With q UE，k+1 The correlation between them has been decreasing, which also leads to the The term is decreasing, leading to a larger theoretical bound In other words, if the rotation angle selected in the rotation model is too small, the positioning accuracy of the R-VLP system will also be reduced. A more understandable angle is that when the rotation angle is too small, it is equivalent to maintaining a stationary state for the UE, and in this state it will fall into the dilemma of traditional VLP being unable to locate in a single LED and single PD scenario. But at the same time, it can be seen that if more rotation postures are selected during the rotation process, the rotation information matrix A can be increased. k , thereby reducing the theoretical limit Ultimately, the R-VLP system achieves higher positioning accuracy.

[0158] In summary, this section derives the CRLB of the R-VLP system model and obtains the influence of various parameters in the rotation model on positioning accuracy.

[0159] 3.3 Particle Swarm Optimization

[0160] In Section 3.2.2 of the present invention, the non-convex optimization problem existing in the rotation positioning problem is mentioned. In order to solve the positioning error problem caused by this non-convex property in the solution, a special non-convex optimization algorithm, namely the particle swarm algorithm, is introduced in this section. The background and basic concepts of the particle swarm algorithm will be introduced in Section 3.3.1, and then the algorithm flow of combining the particle swarm algorithm with R-VLP will be introduced in 3.3.2.

[0161] 3.3.1 Overview of Particle Swarm Optimization

[0162] The inspiration of the particle swarm algorithm comes from the fact that animal groups constantly exchange information with each other during the foraging process. Each individual wants to find the location with the most food. During the search process, they will move in the best direction they feel. Each individual will record the place with the most food they have found during the search process, and share the information within the group after each move. Therefore, from the perspective of the group, each individual has the location with the most food that all individuals in the group have visited. Before the next move, each individual can change the best direction they judge based on the location with the most food they have found and the location with the most food found by the group as a whole, so that after a period of movement, each individual in the group will eventually move to the location where the food is.

[0163] Particle swarm optimization is a group-based intelligent algorithm that has been proven to solve problems such as Figure 4 In the multidimensional non-convex optimization problem in indoor positioning, compared with other intelligent algorithms, the particle swarm algorithm has the advantages of fast convergence speed, few parameters and simple algorithm principle. The most critical thing is that in the multidimensional non-convex problem of indoor positioning, the algorithm that needs to be selected can not only jump out of the local optimal point, but also have a faster convergence speed to achieve higher positioning real-time performance. It is worth mentioning that the particle swarm algorithm cannot completely avoid falling into the local optimal solution. For example, in the above scenario, all individuals in the group are born near a location with the second most food. After continuous movement, the individuals are more likely to concentrate near the suboptimal location. The solution lies in expanding the group size and trying to make the initial position of each individual more uniform and randomly generated in the solution space.

[0164] 3.3.2 Particle Swarm-Based R-VLP Algorithm

[0165] As for the specific algorithm process, the particle swarm algorithm first randomly selects several potential solutions in the solution space, continuously moves each potential solution through multiple iterative processes, and after the iterative process is completed, carefully selects several potential solutions to obtain the final result. The algorithm includes two most important parameters, namely the number of particles N in the particle swarm (the number of potential solutions initially selected) and the number of moves of the particle swarm as a whole M (i.e. the number of iterations of the particle swarm algorithm). In this invention, the particle swarm algorithm is used to solve the positioning optimization problem in the R-VLP system. For the UE position vector p in the optimization problem (3-17), UE The estimation of is equivalent to the estimation of the incident vector r in the rotation positioning model, because in the indoor visible light positioning scene, the position p of the LED light source is LED remains unchanged, so according to formula (3-1), the UE position vector p UE There is a unique mapping relationship between and the incident vector r. For the R-VLP system, each particle in the particle swarm algorithm corresponds to a potential solution to the optimization problem (3-17). n，m To express the incident vector value corresponding to the nth particle (n = 1, 2, ..., N) in the mth iteration process, and use h k，n，m To represent the theoretical channel response value for the k-th posture of the n-th particle in the corresponding rotation process during the m-th iteration.

[0166] First of all, in the initial stage of the particle swarm algorithm, it is necessary to initialize the incident vector r corresponding to N particles n，0 , and give each particle an initial velocity vector v n，0 To complete the subsequent iterative process, the subsequent process is to move (iterate) the entire particle swarm M times, where each iteration includes the following three steps:

[0167] Step 1: Calculate the current fitness of each particle. The objective function of the optimization problem (3-17) can be used to calculate the fitness value ζ of the nth particle. n , the calculation formula is as follows

[0168]

[0169] The fitness value represents the possibility that the position corresponding to the current particle is the final positioning result. In the subsequent process, the movement of the particle can be guided according to the fitness value.

[0170] Step 2: Update the optimal position information of the group and update the particle movement vector. The first step is to select the latest optimal historical position of each particle based on its current fitness value and the historical fitness value in the previous iteration. The process is as follows

[0171]

[0172] in It represents the number of iterations corresponding to the historical optimal position of the nth particle, and the corresponding historical optimal incident vector is In the second step, the historical optimal vector of the entire particle swarm needs to be selected through the historical optimal incident vector of each particle. The process is as follows

[0173]

[0174] in It represents the number of the particle that has reached the historical optimal position of the group in the entire particle swarm, and its corresponding historical optimal incident vector is In the third step, we also need to calculate the moving velocity vector v of the nth particle in the current iteration. n，m , which is calculated as follows:

[0175]

[0176] Among them, c1 and c2 are two non-negative constants, which represent the proportion of the particle's own historical optimal position information and the group's historical optimal position information in the position update process, and r1 and r2 are two random variables that obey a uniform distribution in a closed space from 0 to 1. The meaning of these variables is to give some randomness to the movement of each particle, so as to better avoid falling into the situation of local optimal solution. At the same time, the moving speed v of the nth particle in the current iteration is n，m Will be affected by the maximum movement speed v max The restriction of ||v n，m ||2≤v max , which will ensure that the particle positions in two adjacent iterations will not differ too far, and this will better adapt to the constraints in the optimization problem (3-18).

[0177] Step 3: Movement of particles. Through velocity vector ν n，m , get the incident vector r corresponding to each particle in the next iteration n，m , so far one iteration is completed, and then return to step 1 and repeat.

[0178] Specifically: the incident vector r n，m Update it to r according to the following formula n，m+1

[0179] r n,m =r n,m-1 +v n,m - 1· (3-31)

[0180] In R-VLP, a fixed number of particle swarm iterations M is specified instead of setting a positioning accuracy threshold to terminate the particle swarm iteration process. This makes it easier to analyze the impact of the number of particle swarm iterations on positioning accuracy during the simulation process. After M iterations, the final positioning result is obtained by the incident vector corresponding to the historical optimal position information of the swarm.

[0181]

[0182] The pseudo code of the overall R-VLP positioning algorithm based on the particle swarm algorithm is shown in Table 3-1

[0183] Table 3-1 Pseudocode of the R-VLP positioning algorithm based on particle swarm

[0184]

[0185]

[0186] The time complexity of the algorithm depends on the size N of the particle swarm and the number of particle swarm movements M. In each iteration, the fitness value and speed value of each particle need to be calculated separately, so the time complexity can be expressed as The time complexity of the method of using exhaustive method to obtain positioning results by traversing all possible positions in three-dimensional space is mainly related to the size of the space and the positioning accuracy required by the system. When the size of the space is set to And the required positioning accuracy is When , the time complexity of the exhaustive method is

[0187] For example, when the system positioning accuracy is required to be 0.01 (m) and the space size is 3m*3m*3m, subsequent simulation results show that under the conditions of SNR=50dB, the parameters of the particle swarm algorithm are N=64 and M=100, the R-VLP system can meet this requirement, and in this case, the time complexity of the positioning algorithm of the R-VLP system is 1 / 4000 of the exhaustive method.

[0188] Based on the particle swarm-based R-VLP positioning algorithm proposed in this section, the simulation results of the R-VLP system positioning performance will be presented and analyzed in the next section.

[0189] 3.4 Simulation results

[0190] In this section, we will simulate and verify the R-VLP system, mainly comparing its performance with other single-LED visible light positioning technologies in a single-LED scenario, and the impact of different parameters in the R-VLP system on the system positioning accuracy and the approach to CRLB. The simulation environment adopts an indoor corridor environment, simulating the user's handheld terminal in such a Figure 5 The following figure shows a positioning scenario in a corridor environment. In this scenario, users generally need positioning services to know their location in a certain section of the corridor. However, due to the limited length of the corridor, meter-level positioning accuracy is obviously not applicable. In this case, centimeter-level positioning methods are required to inform users of their current approximate location. This helps users find specific nodes in the corridor, such as corridor murals, cabinets, or closets.

[0191] The default settings of the simulation parameters in this section are shown in Table 3-1, which indicates the default value of a parameter when it is not described in the context of the subsequent simulation diagrams in this section.

[0192] Table 3-1 Simulation experiment parameter settings for positioning in mobile scenarios

[0193]

[0194]

[0195] First, the positioning method proposed in this invention is compared with other single-lamp positioning methods. The simulation results are as follows: Figure 6As shown in the figure, the literature

[56] uses three PD receiving arrays with a certain tilt angle to each other at the receiving end to solve the problem of insufficient number of LED light sources in the scene, which meets the minimum limit of hardware resources for the RSS-based positioning method. The literature

[57] adopts a symmetrical PD receiving array at the receiving end, which includes 5 PDs. In addition to overcoming the problem of insufficient light resources, the system further improves the positioning accuracy compared with the literature

[56] . However, the visible light positioning in the single LED scene proposed in the above two documents has higher requirements for the number of PDs at the receiving end. This requirement cannot be met on general terminal devices, and the special structure of this receiving end has a stronger restriction on the incident angle of the PD end. This means that at some edge positions where the LED can be illuminated, the incident angle of some PDs with special angles in the receiving array exceeds its FOV, which ultimately leads to positioning failure. Compared with these two positioning systems, the proposed R-VLP positioning system not only requires only one PD on the receiving device at the receiving end, but also has a lower restriction on the incident angle on the PD end. This means that in the scenario of only a single LED, the R-VLP system can complete positioning in a larger range. It can be seen that when K = 3 alone, the positioning performance has exceeded the above two systems. As the number of rotation postures increases to K = 6, it can be seen that the positioning performance of the R-VLP system has approached the positioning accuracy of the RSS-based positioning system in the traditional three-lamp scenario.

[0196] Figure 7 The positioning accuracy of the R-VLP system at different positions on different planes is shown. It can be seen that at height z UE = 1, the positioning accuracy at any position on the plane is worse than that at height z UE =2, while on the plane at the same height, its positioning accuracy shows a trend of decreasing from the middle position to the edge position of the plane. When SNR = 60dB, it can be seen that any position on the two planes can achieve a positioning accuracy of 0.1m. This means that for applications with centimeter-level accuracy requirements, R-VLP can meet its requirements in the case of a single lamp with this parameter setting. The simulation results also verify the relationship between the CRLB and distance of the R-VLP system derived in Section 3.3.3, that is, in the R-VLP system, the positioning accuracy is always negatively correlated with the square of the distance d between the UE and the LED. From the article

[55] that explores the performance limitations of indoor visible light systems for LED positioning, it can be seen that no matter what positioning principle the VLP system is based on, its positioning accuracy will decrease with increasing distance.

[0197] Figure 8The relationship between the positioning accuracy of the R-VLP system and the number of particles N in the particle swarm algorithm is shown. It can be seen that as the number of particles N increases, the positioning accuracy of the system increases accordingly. This means that as the size of the particle swarm increases, the incident vectors corresponding to the initially generated particles will be more randomly distributed in the solution space, which will also reduce the probability of the algorithm falling into a local optimal solution. However, a larger particle swarm size not only means that higher positioning accuracy can be achieved, but also means that the time complexity of the positioning system's solution algorithm increases, resulting in reduced real-time positioning. It is worth mentioning that, whether in the case of SNR = 40dB or SNR = 60dB, when the particle swarm size N approaches 128, the performance of the R-VLP system in simulation approaches its CRLB.

[0198] Figure 9 The relationship between the positioning accuracy of the R-VLP system and the number of particle moves M in the particle swarm algorithm is shown. It can be seen that as the number of particle swarm moves M increases, the positioning accuracy of the system increases accordingly. This means that when the particle swarm moves more times, the particles can eventually reach a position closer to the final positioning result. However, compared with the particle swarm size N, when the number of particle swarm moves M is relatively low, the system's positioning accuracy will be worse. However, as it increases, the system's positioning effect will improve rapidly, and the time complexity of the positioning algorithm will increase. However, this also means that when the system has real-time and accuracy constraints, it is possible to prioritize adjusting the particle swarm size M to make the R-VLP system meet its requirements. It is worth mentioning that whether in the case of SNR = 40dB or SNR = 60dB, when the number of particle swarm moves M approaches 100, the R-VLP system will also approach its CRLB.

[0199] Figure 10 The approximation effect of the R-VLP system on its CRLB is shown under different combinations of particle swarm parameters. It can be seen that when the particle swarm size N = 128 and the number of particle swarm movements M = 100, the R-VLP system can achieve a good approximation effect on its CRLB, which also guides the default values of the particle swarm parameters proposed at the beginning of this invention. Figure 10 The simulation also verified the positioning effect when the incident angle and the rotation angle are not constant during the rotation process of the system. This means that in practical applications, the positioning process can be completed without special restrictions on these two angle parameters, which also shows the robustness of the R-VLP system to angle parameters.

[0200] Figure 11The influence of the number of rotational postures in the rotation model on the positioning accuracy of the R-VLP system is explained. It can be seen that the overall trend is that as the number of rotational postures K increases, the system positioning accuracy will also increase, but the growth trend will continue to decrease, which also means that the system accuracy cannot be unlimitedly improved by unlimitedly increasing the number of postures. Figure 12 and Figure 13 The deflection angle Ψ in the rotation model is respectively described k and the rotation angle Υ k，k+1 Regarding the impact on system positioning accuracy, the overall trend is that as the deflection angle increases, the system positioning accuracy increases accordingly, but please note that the deflection angle cannot exceed the receiver's FOV (π / 3 in this invention). Similarly, as the rotation angle increases, the system positioning accuracy also increases, but the rotation angle cannot be selected as π, otherwise two adjacent rotation postures will be completely correlated and no more rotation information will be provided.

[0201] The present invention mainly studies the problem of static visible light positioning in a single LED scenario. First, the rotation model proposed for the first time is introduced, and its characteristics are compared with the existing VLP system under a single LED. Then the R-VLP positioning system combined with the rotation model is explained. The key point is to transform the existing Rembrandt channel model so that it can provide a set of nonlinear equations that are clearly related to the angle and position information during the rotation process. Then, an optimization problem is formed to solve the positioning result, and its non-convexity is explained. In order to explore its theoretical positioning performance, the CRLB of the system positioning accuracy is derived, and the influence of different system parameters on the system performance is explained. Subsequently, in order to solve the non-convexity of the positioning optimization problem, the particle swarm algorithm is introduced and introduced, and adapted to the rotation positioning model. At the end of the present invention, the performance of the R-VLP system under different particle swarm parameters and rotation model parameters, as well as the situation approaching the theoretical limit, are simulated and verified. At the same time, the positioning effects are compared with other single-LED VLP systems and traditional RSS-based VLP systems. The final results show that the R-VLP system is not only superior to the existing VLP system in the single-LED scenario, but also its performance will approach or even exceed that of the original RSS-based VLP system as the number of rotation postures increases.

[0202] The beneficial effects of the present invention are: 1. The method of the present invention can be used in indoor positioning scenarios of ISAC; 2. The method of the present invention allows a UE equipped with a single PD to rotate multiple times in a single-lamp situation to complete the positioning process; 3. The method of the present invention is robust to the number of hardware devices and can complete the work in any indoor scenario. In this respect, it is far superior to existing methods that require at least three LEDs or three PDs, such as: trilateration method, fingerprint method and approximation method.

[0203] The above is a further detailed description of the present invention in conjunction with specific preferred embodiments, and the specific implementation of the present invention should not be considered to be limited to these descriptions. For those skilled in the art of the present invention, without departing from the concept of the present invention, several simple deductions or substitutions can be made, which should be considered to fall within the scope of protection of the present invention.

Claims

1. A single-transmit and single-receive three-dimensional visible light positioning method based on rotation, characterized in that: include: Step 1, positioning scene parameterization step; use x min with x max Indicates the minimum and maximum coordinates that the UE can reach in the horizontal direction in the scene, y min with y max Indicates the minimum and maximum coordinates that the UE can reach in the vertical direction in the scene, z min With z max Indicates the minimum and maximum coordinates that the UE can reach in the height direction of the scene; Step 2, rotation positioning step: Use the channel responses under different postures obtained during the terminal rotation process to perform the positioning process, that is, find a point in space where the Euclidean distance between its theoretical channel response value under K postures and the channel response value actually obtained through channel estimation during the rotation process is minimized; Step 3: Solve the non-convexity problem; introduce the particle swarm optimization algorithm to solve the positioning error problem caused by the non-convexity in the positioning process of step 2 and obtain the final positioning result.

2. The single-transmit and single-receive three-dimensional visible light positioning method according to claim 1, characterized in that: In step 2, the positioning process is regarded as an optimization problem, which is expressed as: In the formula ——The final positioning result of the R-VLP system; ——Any point in the solution space; ——The theoretical channel response value at any point in the kth posture; h k ——The channel response value under the k-th posture estimated by the channel during the actual rotation process; q LED ——Indicates the normal vector of the LED; r——the incident vector between the transmitting end LED and the receiving end PD; The constraint C1 in formula (3-17) indicates that the possible positioning results should exist within the coverage range of the LED light, that is, the final positioning search range is a cone range from the LED downward. The constraints C2-C5 indicate that the search should be within the set fixed range, and then through the limited range x min with x max 、y min with y max and z min With z max To search.

3. The single-transmit and single-receive three-dimensional visible light positioning method according to claim 2, characterized in that: The specific steps of step 3 include: Step 30: Input step: Input the channel estimation results under K rotation postures Where k = 0, 1, ..., K, particle swarm size N and particle swarm iteration number M; Step 31, initialization step: In the initial stage of the particle swarm algorithm, it is necessary to initialize the incident vector r corresponding to N particles n，0 , and give each particle an initial velocity vector v n，0 To complete the subsequent iterative process, the subsequent process is to iterate the entire particle swarm M times; Step 32, the positioning result output step; the incident vector corresponding to the historical optimal particle of the group represents the historical optimal position of the group Output as the final positioning result of R-VLP.

4. The single-transmit and single-receive three-dimensional visible light positioning method according to claim 3, characterized in that: The step 31 specifically includes: Step 1: Calculate the current fitness value ζ of each particle n ; Step 2: Based on the latest fitness value ζ n , update the optimal position information of the group and update the particle movement speed vector v n，m ; Step 3, through the velocity vector v n，m , get the incident vector r corresponding to each particle in the next iteration n，m , so far one iteration is completed, and then return to step 1 and repeat.

5. The single-transmit and single-receive three-dimensional visible light positioning method according to claim 4, characterized in that: In step 1, the objective function of the optimization problem (3-17) is used to calculate the fitness value of the nth particle, and its calculation formula is as follows: The fitness value represents the possibility that the position corresponding to the current particle is the final positioning result. In the subsequent process, the movement of the particle is guided by the fitness value.

6. The single-transmit and single-receive three-dimensional visible light positioning method according to claim 4, characterized in that: The second step includes: In the first step, the latest optimal historical position of each particle is selected based on its current fitness value and its historical fitness value in the previous iteration. The process is as follows: in It represents the number of iterations corresponding to the historical optimal position of the nth particle, and the corresponding historical optimal incident vector is In the second step, the historical optimal vector of the entire particle swarm is selected through the historical optimal incident vector of each particle. The process is as follows: in It represents the number of the particle that has reached the historical optimal position of the group in the entire particle swarm, and its corresponding historical optimal incident vector is Step 3: Calculate the moving velocity vector v of the nth particle in the current iteration n，m , the calculation formula is as follows: Among them, c1 and c2 are two non-negative constants, which represent the proportion of the particle's own historical optimal position information and the group's historical optimal position information in the position update process. r1 and r2 represent two random variables that obey a uniform distribution in the closed space from 0 to 1.

7. The single-transmit and single-receive three-dimensional visible light positioning method according to claim 6, characterized in that: The step three is specifically as follows: The incident vector r n，m Update it to r according to the following formula n，m+1 r n,m ＝r n，m-1 +v n,m-1 (3-31)。 8. The single-transmit and single-receive three-dimensional visible light positioning method according to claim 3, characterized in that: In step 32, a fixed number of particle swarm iterations M is specified in the R-VLP. After M iterations, the final positioning result is obtained by the incident vector corresponding to the historical optimal position information of the swarm. The formula is as follows:

Images