Nonlinear hybrid beam forming method and system in unmanned aerial vehicle communication, and storage medium

Abstract

The invention discloses a nonlinear hybrid beam forming method and system in unmanned aerial vehicle communication and a storage medium. The method comprises the following steps: calculating a constraint condition of a hybrid precoding matrix of a transmitting end; constructing a target function of an analog precoding matrix of the transmitting end according to constraint conditions of the hybrid precoding matrix of the transmitting end and requirements of an optimal precoder; the optimal gradient iteration step number is calculated according to the relation between the historical gradient information and the time delay / or the spectrum efficiency and / or the energy efficiency; calculating an optimal solution of a transmitting end simulation precoding matrix objective function according to the optimal gradient iteration step number; obtaining a transmitting end hybrid precoding matrix according to the optimal solution of the transmitting end analog precoding matrix objective function; and performing hybrid beam forming according to the hybrid precoding matrix of the transmitting end. The method solves the problem that the loading capacity requirement and the real-time requirement of unmanned aerial vehicle communication cannot be met due to the fact that an existing millimeter wave hybrid beam forming technology is insufficient in spectrum efficiency, high in time delay and low in energy efficiency.

Description

Technical Field

[0001] This invention belongs to the field of communication technology, and in particular relates to a nonlinear hybrid beamforming method, system and storage medium for UAV communication. Background Technology

[0002] In low-altitude communication scenarios such as drone inspection, drone live streaming, and drone delivery, high-speed, high-bandwidth communication systems are required between drones and ground control centers to achieve accurate fault location and high-definition image transmission. Millimeter waves provide abundant high-frequency bandwidth resources, which can significantly improve drones' real-time video transmission, flight status monitoring, high-precision positioning, and remote control capabilities. However, millimeter waves have characteristics such as short wavelength, high propagation path loss, high penetration loss, difficulty in diffraction, and high energy consumption, which are not conducive to long-distance communication.

[0003] With the development of millimeter-wave hybrid beamforming technology, the combination of large-scale antenna arrays for UAVs with millimeter waves can generate narrow beams with high gain, while also overcoming path loss and attenuation problems during millimeter-wave propagation, thus improving the system's spectrum reuse rate and spectral efficiency. However, existing hybrid beamforming technologies suffer from insufficient spectral efficiency due to the need to solve and store the inverse matrix for each iteration. Furthermore, they exhibit high computational complexity, high computational latency, and low energy efficiency, failing to meet the payload capacity and real-time requirements of small-volume UAVs.

[0004] To address the issues of insufficient spectral efficiency, high latency, and low energy efficiency in existing millimeter-wave hybrid beamforming technologies for UAV communication, which prevent them from meeting the payload capacity and real-time requirements of UAV communication, a nonlinear hybrid beamforming method, system, and storage medium for UAV communication are proposed. Summary of the Invention

[0005] This invention proposes a nonlinear hybrid beamforming method, system, and storage medium for UAV communication, which at least solves the problems of insufficient spectral efficiency, high latency, and low energy efficiency of existing millimeter-wave hybrid beamforming technology, which makes it unable to meet the payload capacity and real-time requirements of UAV communication.

[0006] According to an embodiment of the present invention, a nonlinear hybrid beamforming method for unmanned aerial vehicle (UAV) communication is provided, comprising the following steps:

[0007] Calculate the constraints of the transmitter hybrid precoding matrix based on the UAV communication channel status and RF antenna status;

[0008] The objective function of the transmitter's analog precoding matrix is ​​constructed based on the constraints of the transmitter's hybrid precoding matrix and the requirements of the optimal precoder.

[0009] The optimal gradient iteration steps are calculated based on the relationship between historical gradient information and time delay / or spectral efficiency and / or energy efficiency.

[0010] The optimal solution of the objective function of the transmitter's simulated precoding matrix is ​​calculated based on the optimal gradient iteration steps;

[0011] The transmitter hybrid precoding matrix is ​​obtained based on the optimal solution of the objective function of the transmitter simulated precoding matrix;

[0012] Hybrid beamforming is performed based on the hybrid precoding matrix at the transmitter.

[0013] In a preferred embodiment, the constraint conditions for calculating the transmitter hybrid precoding matrix based on the UAV communication channel state and radio frequency antenna state include:

[0014] The power constraint constant of the UAV transmitter is calculated based on the UAV communication channel state and RF antenna state, denoted as . ;

[0015] The power constraint condition for the transmitter is calculated based on the Frobenius norm. ,in, This represents the analog precoding matrix at the transmitting end. This represents the digital precoding matrix at the transmitting end. This represents the transmitter hybrid precoding matrix; the transmitter hybrid precoding matrix is ​​obtained by multiplying the transmitter's digital precoding matrix and the transmitter's analog precoding matrix.

[0016] In a preferred embodiment, the step of constructing the objective function of the transmitter analog precoding matrix based on the constraints of the transmitter hybrid precoding matrix and the requirements of the optimal precoder includes the following steps:

[0017] The optimal digital precoding matrix for all-digital beamforming is obtained based on the singular value decomposition of the channel matrix. ;

[0018] The initial objective function is constructed by minimizing the Euclidean distance between the optimal digital precoding matrix and the transmitter hybrid precoding matrix, and is expressed as: ;

[0019] For the initial objective function Taking the partial derivative yields ,in yes Moore-Penrose pseudo-inverse;

[0020] The objective function for constructing the transmitter simulation precoding matrix is ​​based on the partial derivative of the initial objective function and the Frobenius norm, and is expressed as follows: , , It is the set of feasible solutions for an analog pre-encoder with constant modulus constraints.

[0021] In a preferred embodiment, the step of calculating the optimal gradient iteration steps based on the relationship between historical gradient information and time delay / or spectral efficiency and / or energy efficiency includes the following steps:

[0022] The functional relationship between latency and the number of steps of historical gradient information at iteration points is calculated based on the relationship between the complexity of the non-monotonic linear search algorithm and the historical gradient information at iteration points.

[0023] The spectral efficiency is fitted and calculated as a function of the number of steps of historical gradient information at different iteration points to determine the spectral efficiency.

[0024] The functional relationship between energy efficiency and the number of steps of historical gradient information at different iteration points is calculated by fitting the energy efficiency corresponding to the number of steps of historical gradient information at different iteration points.

[0025] An optimization model for the number of iteration point historical gradient information steps is constructed based on the functional relationship between time delay and / or spectral efficiency and / or energy efficiency and the number of iteration point historical gradient information steps.

[0026] The optimal gradient iteration steps are calculated based on the historical gradient information steps corresponding to the maximum value of the iteration point optimization model.

[0027] In a preferred embodiment, the step of calculating the functional relationship between the time delay and the number of steps of the historical gradient information at the iteration point based on the relationship between the complexity of the non-monotonic linear search algorithm and the historical gradient information at the iteration point includes the following steps:

[0028] Calculate the inherent computational complexity of non-monotonic linear search algorithms; the inherent computational complexity includes the computational complexity of Euclidean gradient, Riemann gradient, line search, orthogonal projection, and shrinkage.

[0029] The additional computational complexity of the non-monotonic linear search algorithm is calculated based on the functional relationship between the number of steps and the step size in historical gradient information.

[0030] The complexity of the non-monotonic linear search algorithm is obtained based on its inherent computational complexity and the additional computational complexity of the non-monotonic linear search algorithm.

[0031] The functional relationship between latency and the number of historical gradient information steps at different iteration points is calculated based on the impact of the number of iteration points on the complexity of the non-monotonic linear search algorithm.

[0032] In a preferred embodiment, the step of calculating the optimal solution of the objective function of the transmitter's simulated precoding matrix based on the optimal gradient iteration step number includes the following steps:

[0033] Step 401: Construct the Riemannian complex circular manifold and initialize the iteration points (initial step size).

[0034] Step 402: Calculate the cost function and Riemann gradient at the iteration point;

[0035] Step 403: Calculate the optimal step size for the current iteration point based on the optimal gradient iteration step number;

[0036] Step 404: Calculate the optimal search direction for the current iteration point based on the Riemann gradient and the optimal step size for the current iteration point;

[0037] Step 405: Using Armijo criterion-based line search, search in the tangent space according to the optimal search direction of the current iteration point and shrink back to the manifold to obtain the new iteration point;

[0038] Step 406: Store the gradient information of the current iteration point and update the optimal step size of the current iteration point;

[0039] Step 407: Determine whether the current iteration point meets the termination condition. If yes, stop the iteration counting and obtain the optimal solution of the objective function of the transmitter's simulated precoding matrix; otherwise, return to step 402. The termination condition is that the norm of the gradient reaches a preset value and the optimal step size of the current iteration point is less than the preset minimum step size.

[0040] In a preferred embodiment, obtaining the transmitter hybrid precoding matrix based on the optimal solution of the objective function of the transmitter analog precoding matrix includes the following steps:

[0041] The transmitter-simulated precoding matrix is ​​obtained from the optimal solution of the objective function of the transmitter-simulated precoding matrix;

[0042] Calculate the transmitter's digital precoding matrix based on the optimal digital precoding matrix and the analog precoding matrix;

[0043] The transmitter hybrid precoding matrix is ​​calculated based on the transmitter analog precoding matrix and the transmitter digital precoding matrix.

[0044] In a preferred embodiment, the hybrid beamforming based on the transmitter hybrid precoding matrix includes:

[0045] The amplitude and phase of the transmitted signal are adjusted according to the digital precoding matrix at the transmitter.

[0046] A phase shifter is constructed based on the transmitter's simulated precoding matrix and connected to the radio frequency link;

[0047] The transmitted signal, after amplitude and phase adjustment and phase shifting, is transmitted through a radio frequency antenna to form a hybrid beam.

[0048] According to another embodiment of the present invention, a computer-readable storage medium is provided that stores a computer program for electronic data interchange, wherein the computer program causes a computer to execute the nonlinear hybrid beamforming method described above in UAV communication.

[0049] According to another embodiment of the present invention, a nonlinear hybrid beamforming system for unmanned aerial vehicle (UAV) communication is provided, comprising:

[0050] processor;

[0051] Memory;

[0052] And one or more programs, wherein the one or more programs are stored in memory and configured to be executed by the signal processing unit, the programs causing the computer to perform the nonlinear hybrid beamforming method in the above-described UAV communication.

[0053] The advantages of the nonlinear hybrid beamforming method, system, and storage medium in UAV communication of the present invention are:

[0054] (1) The function relationship between time delay and the number of steps of historical gradient information at different iteration points is calculated based on the influence of the number of steps of historical gradient information at different iteration points on the complexity of non-monotonic linear search algorithm. Compared with the traditional nonlinear hybrid beamforming algorithm, it can effectively identify the different effects of different step lengths on time delay, which makes it easier to optimize the step length from the perspective of time delay, thereby improving the efficiency of hybrid beamforming algorithm.

[0055] (2) The spectral efficiency is fitted and calculated according to the spectral efficiency corresponding to the number of steps of the historical gradient information at different iteration points. Compared with the traditional nonlinear hybrid beamforming algorithm, it can effectively identify the different effects of different step lengths on spectral efficiency, which makes it easier to optimize the step length from the perspective of spectral efficiency, thereby improving the frequency efficiency of hybrid beamforming.

[0056] (3) The energy efficiency is fitted and calculated according to the energy efficiency corresponding to the number of steps of the historical gradient information at different iteration points. Compared with the traditional nonlinear hybrid beamforming algorithm, it can effectively identify the different effects of different step lengths on energy efficiency, which makes it easier to optimize the step length from the perspective of energy efficiency, thereby improving the energy efficiency of hybrid beamforming.

[0057] (4) Based on the reconstructed objective function, the Riemann gradient of the objective function is calculated by constructing the Riemann complex circular manifold. The constant modulus constraint of the simulation domain is transformed into unconstrained optimization. When constructing the search direction, the RBB method only requires the current step size and the Riemann gradient. Compared with the traditional conjugate gradient method or the traditional quasi-Newton method, which requires the calculation of complex matrices in each iteration, it can perform efficient search and converge faster in the highly constrained feasible domain, effectively reducing computational complexity and computational delay.

[0058] (5) Calculate the optimal gradient iteration step based on the relationship between historical gradient information and time delay / or spectral efficiency and / or energy efficiency, and calculate the step size and search direction based on the optimal gradient iteration step. Compared with the traditional RBB algorithm and hybrid beamforming algorithm, it can not only make the line search conditions relatively relaxed and avoid getting trapped in local optima, but also effectively reduce the number of algorithm iterations, improve the algorithm efficiency while ensuring the frequency efficiency and energy efficiency of the algorithm, and improve the adaptability of the algorithm in UAV communication scenarios. Attached Figure Description

[0059] Figure 1 This is a flowchart of a nonlinear hybrid beamforming method in UAV communication according to an embodiment of the present invention.

[0060] Figure 2 This is a flowchart of step S01 of the nonlinear hybrid beamforming method in UAV communication according to an embodiment of the present invention.

[0061] Figure 3 This is a flowchart of step S02 of the nonlinear hybrid beamforming method in UAV communication according to an embodiment of the present invention.

[0062] Figure 4 This is a flowchart of step S03 of the nonlinear hybrid beamforming method in UAV communication according to an embodiment of the present invention.

[0063] Figure 5 This is a flowchart of sub-step S04 of the nonlinear hybrid beamforming method in UAV communication according to an embodiment of the present invention.

[0064] Figure 6 This is a flowchart of sub-step S06 of the nonlinear hybrid beamforming method in UAV communication according to an embodiment of the present invention.

[0065] Figure 7 This is a flowchart of sub-step S07 of the nonlinear hybrid beamforming method in UAV communication according to an embodiment of the present invention.

[0066] Figure 8 This is a flowchart of sub-step S08 of the nonlinear hybrid beamforming method in UAV communication according to an embodiment of the present invention.

[0067] Figure 9 This is a schematic diagram of the nonlinear hybrid beamforming system in UAV communication according to an embodiment of the present invention.

[0068] Figure 10 The graph shows the time delay versus signal-to-noise ratio for different beamforming algorithms.

[0069] Figure 11 This is a schematic diagram of a nonlinear hybrid beamforming system for UAV communication. Detailed Implementation

[0070] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0071] According to an embodiment of the present invention, a nonlinear hybrid beamforming method for UAV communication is provided, the flowchart of which is shown below. Figure 1 As shown, it includes:

[0072] Step S01: Calculate the constraints of the transmitter hybrid precoding matrix based on the UAV communication channel status and RF antenna status;

[0073] Step S02: Construct the objective function of the transmitter's analog precoding matrix based on the constraints of the transmitter's hybrid precoding matrix and the requirements of the optimal precoder;

[0074] Step S03: Calculate the optimal number of gradient iteration steps based on the relationship between historical gradient information and time delay / or spectral efficiency and / or energy efficiency;

[0075] Step S04: Calculate the optimal solution of the objective function of the transmitter's simulated precoding matrix based on the optimal gradient iteration step number;

[0076] Step S05: Obtain the transmitter hybrid precoding matrix based on the optimal solution of the objective function of the transmitter's simulated precoding matrix;

[0077] Step S06: Perform hybrid beamforming based on the hybrid precoding matrix of the transmitter.

[0078] In a preferred embodiment, step S01, calculating the constraints of the transmitter hybrid precoding matrix based on the UAV communication channel state and RF antenna state, is illustrated in the flowchart below. Figure 2 As shown, it includes:

[0079] Step S011: Calculate the power constraint constant of the UAV transmitter based on the UAV communication channel status and radio frequency antenna status;

[0080] Step S012: Calculate the power constraint conditions of the transmitter based on the Frobenius norm.

[0081] In this embodiment, the transmitter power constraint constants corresponding to different channel states and radio frequency antenna states are preset according to the channel states under various drone usage scenarios, as well as the number of antennas and radio frequency power of the drone.

[0082] The power constraint constant of the UAV transmitter is obtained based on the current UAV communication channel status and RF antenna status, denoted as: The transmitter radio frequency link is denoted as The number of radio frequency antennas is denoted as The raw data stream is represented as After passing through the dimension of Digital baseband beamformer and dimension for The simulated radio frequency beamformer yields a dimension of signal The normalized transmit power constraint is set to The digital precoding matrix at the transmitter corresponding to the digital baseband beamformer. The analog precoding matrix of the transmitter corresponding to the analog radio frequency beamformer The transmitter's hybrid precoding matrix is ​​obtained by multiplying the transmitter's digital precoding matrix and analog precoding matrix. The power constraint condition for the transmitter is calculated based on the Frobenius norm. Among them, the constant modulus constraint representation of the simulated precoding matrix , It is the set of feasible solutions for an analog pre-encoder with constant modulus constraints.

[0083] In a preferred embodiment, step S02 involves constructing the objective function of the transmitter's analog precoding matrix based on the constraints of the transmitter's hybrid precoding matrix and the requirements of the optimal precoder. The flowchart is shown below. Figure 3 As shown, the steps include:

[0084] Step S021: Obtain the optimal digital precoding matrix for all-digital beamforming based on the singular value decomposition of the channel matrix;

[0085] Step S022: Construct the initial objective function based on minimizing the Euclidean distance between the optimal digital precoding matrix and the transmitter hybrid precoding matrix;

[0086] Step S023: Obtain the least squares solution by taking the partial derivative of the initial objective function;

[0087] Step S024: Construct the objective function of the transmitter simulation precoding matrix based on the partial derivative solution of the initial objective function and the Frobenius norm.

[0088] In this embodiment, according to the channel matrix Singular value decomposition yields the optimal digital precoding matrix for all-digital beamforming, denoted as . The initial objective function is constructed by minimizing the Euclidean distance between the optimal digital precoding matrix and the transmitter hybrid precoding matrix, and is expressed as: For the initial objective function pair Taking the partial derivative, we obtain the least squares solution. ,in yes The Moore-Penrose pseudoinverse; the objective function of the transmitter-simulated precoding matrix is ​​constructed based on the least squares solution and the Frobenius norm, expressed as: , , It is the set of feasible solutions for an analog pre-encoder with constant modulus constraints.

[0089] In a preferred embodiment, step S03, calculating the optimal gradient iteration steps based on the relationship between historical gradient information and time delay / or spectral efficiency and / or energy efficiency, is illustrated in the flowchart below. Figure 4 As shown, the steps include:

[0090] Step S031: Calculate the functional relationship between the time delay and the number of steps of the historical gradient information at the iteration point based on the relationship between the complexity of the non-monotonic linear search algorithm and the historical gradient information at the iteration point;

[0091] Step S032: Fit and calculate the functional relationship between spectral efficiency and the number of steps of historical gradient information at different iteration points based on the spectral efficiency corresponding to the number of steps of historical gradient information at different iteration points;

[0092] Step S033: Fit and calculate the functional relationship between energy efficiency and the number of steps of historical gradient information at different iteration points based on the energy efficiency corresponding to the number of steps of historical gradient information at different iteration points;

[0093] Step S034: Construct an optimization model for the number of iteration points based on the functional relationship between time delay and / or spectral efficiency and / or energy efficiency and the number of iteration points in the historical gradient information.

[0094] Step S035: Calculate the optimal gradient iteration steps based on the historical gradient information steps corresponding to the maximum value of the iteration point in the optimization model.

[0095] In a preferred embodiment, step S031 involves calculating the functional relationship between the delay and the number of steps for the historical gradient information at the iteration point, based on the relationship between the complexity of the non-monotonic linear search algorithm and the historical gradient information at the iteration point. The flowchart is as follows: Figure 5 As shown, the steps include:

[0096] Step S0311: Calculate the inherent computational complexity of the non-monotonic linear search algorithm; the inherent computational complexity includes the computational complexity of Euclidean gradient, Riemann gradient, line search, orthogonal projection, and shrinkage.

[0097] Step S0312: Calculate the new computational complexity of the non-monotonic linear search algorithm based on the functional relationship between the number of steps and the step size in the historical gradient information;

[0098] Step S0313: Obtain the complexity of the non-monotonic linear search algorithm based on its inherent computational complexity and the additional computational complexity of the non-monotonic linear search algorithm.

[0099] Step S0314: Calculate the functional relationship between the delay and the number of iteration points based on the impact of the number of historical gradient information steps at different iteration points on the complexity of the non-monotonic linear search algorithm.

[0100] In this embodiment, the inherent computational complexity of the non-monotonic linear search algorithm includes the computational complexity of Euclidean gradient, Riemann gradient, orthogonal projection, shrinking, and line search. It is assumed that... , And it indicates the number of receiving antennas, The complexity of calculating the Euclidean gradient is O(n log n). The complexity of calculating the Riemann gradient is O(n). The computational complexity of orthogonal projection is O(n). The computational complexity of shrinking is The computational complexity of linear search is The sum of the computational complexities of each algorithm is the inherent computational complexity of the non-monotonic linear search algorithm. .

[0101] Use variables The step size represents the number of steps for the historical gradient information at each iteration point. This indicates that the increased computational complexity of the non-monotonic linear search algorithm is calculated based on the functional relationship between the number of iterations and the step size in the iteration point historical gradient information. .

[0102] The complexity of the non-monotonic linear search algorithm is obtained based on its inherent computational complexity and the additional computational complexity. .

[0103] The calculation of the time delay as a function of the number of iterations based on the historical gradient information steps at different iteration points on the complexity of the non-monotonic linear search algorithm is achieved by training with different iterations based on the historical gradient information steps. The corresponding non-monotonic linear search algorithm complexity The change in time delay is used to obtain a functional relationship between the time delay and the number of iterations based on historical gradient information. The time delay is expressed as a variable. If we express this as a function of the time delay and the number of iterations based on historical gradient information, then the relationship between the time delay and the number of iterations is expressed as follows: .

[0104] In step S032, the step of fitting and calculating the functional relationship between spectral efficiency and the number of iteration points' historical gradient information steps based on the spectral efficiency corresponding to the number of iteration points' historical gradient information steps is achieved by training the number of iteration points' historical gradient information steps. The change in the spectral efficiency of the corresponding hybrid beamforming yields a functional relationship between spectral efficiency and the number of iterations using historical gradient information. The spectral efficiency is expressed as a variable... The functional relationship between spectral efficiency and the number of iterations of historical gradient information is expressed as: .

[0105] In step S033, the step of fitting and calculating the functional relationship between energy efficiency and the number of iteration points' historical gradient information steps based on the energy efficiency corresponding to the number of iteration points' historical gradient information steps is achieved by training the number of iteration points' historical gradient information steps. The change in energy efficiency of the corresponding hybrid beamforming yields a functional relationship between energy efficiency and the number of iteration steps based on historical gradient information. Energy efficiency is expressed as a variable. If we express this as a function of the number of iterations for the historical gradient information at each iteration point, then the relationship between energy efficiency and the number of iterations is expressed as follows: .

[0106] In step S034, the step of constructing the iteration point historical gradient information step count optimization model based on the functional relationship between time delay and / or spectral efficiency and / or energy efficiency and the number of iteration point historical gradient information steps is calculated based on the positive correlation between the iteration point historical gradient information step count optimization model and the functional relationship between time delay and the number of iteration point historical gradient information steps, and / or the functional relationship between spectral efficiency and the number of iteration point historical gradient information steps, and / or the functional relationship between energy efficiency and the number of iteration point historical gradient information steps. The iteration point historical gradient information step count optimization model is a functional relationship between the algorithm's overall efficiency value and the number of iteration point historical gradient information steps. The algorithm's overall efficiency value is represented by a variable. The iteration point historical gradient information step optimization model is expressed as follows: .

[0107] Examples A1-A7 illustrate different implementations of the step optimization model for calculating historical gradient information at iteration points, as follows:

[0108] Example A1: Calculate the optimization model for the number of iteration points based on the functional relationship between time delay and the number of iteration point historical gradient information steps and the positive correlation between the optimization model for the number of iteration point historical gradient information steps.

[0109] Specifically, the functional relationship between time delay and the number of iterations for historical gradient information is expressed as: The optimization model for the number of iteration points based on the functional relationship between time delay and the number of iteration points using historical gradient information is calculated. In a preferred embodiment, the step optimization model calculates the historical gradient information at iteration points. ,in , , These are the calculated coefficients obtained through prior training.

[0110] Example A2: Calculate the optimization model for the number of iteration points based on the functional relationship between spectral efficiency and the number of iteration point historical gradient information steps and the positive correlation between the optimization model for the number of iteration point historical gradient information steps.

[0111] Specifically, the functional relationship between spectral efficiency and the number of iterations for historical gradient information is expressed as: The optimization model for the number of iteration points based on the functional relationship between spectral efficiency and the number of iteration point historical gradient information steps, and the positive correlation between these relationships and the optimization model for the number of iteration point historical gradient information steps, is calculated. In a preferred embodiment, the step optimization model calculates the historical gradient information at iteration points. ,in , , These are the calculated coefficients obtained through prior training.

[0112] Example A3: Calculate the optimization model for the number of iteration points based on the functional relationship between energy efficiency and the number of iteration point historical gradient information steps, and the positive correlation between the optimization model for the number of iteration point historical gradient information steps.

[0113] Specifically, the functional relationship between energy efficiency and the number of iterations for historical gradient information is expressed as: The optimization model based on the functional relationship between energy efficiency and the number of iterations using historical gradient information is calculated. In a preferred embodiment, the step optimization model calculates the historical gradient information at iteration points. ,in , , These are the calculated coefficients obtained through prior training.

[0114] Example A4: Calculate the optimization model for the number of iteration points based on the functional relationship between time delay and the number of iteration point historical gradient information steps, the functional relationship between spectral efficiency and the number of iteration point historical gradient information steps, and the positive correlation between these relationships and the optimization model for the number of iteration point historical gradient information steps.

[0115] Specifically, the functional relationship between time delay and the number of iterations for historical gradient information is expressed as: The functional relationship between spectral efficiency and the number of iterations for historical gradient information is expressed as follows: Based on the functional relationship between time delay and the number of iteration point historical gradient information steps, and the positive correlation between the functional relationship between spectral efficiency and the number of iteration point historical gradient information steps and the optimization model of the number of iteration point historical gradient information steps, the optimization model of the number of iteration point historical gradient information steps is calculated. In a preferred embodiment, the step optimization model calculates the historical gradient information at iteration points. ,in , , , , These are the calculated coefficients obtained through prior training. In another preferred embodiment, the step optimization model calculates the historical gradient information at iteration points. ,in , , , These are the calculated coefficients obtained through prior training.

[0116] Example A5: Calculate the optimization model for the number of iteration points based on the functional relationship between time delay and the number of iteration point historical gradient information steps, the functional relationship between energy efficiency and the number of iteration point historical gradient information steps, and the positive correlation between the optimization model for the number of iteration point historical gradient information steps.

[0117] Specifically, the functional relationship between time delay and the number of iterations for historical gradient information is expressed as: The functional relationship between energy efficiency and the number of iterations for historical gradient information is expressed as follows: Based on the functional relationship between time delay and the number of iterations using historical gradient information, and the positive correlation between energy efficiency and the number of iterations using historical gradient information, the optimization model for the number of iterations using historical gradient information is calculated. In a preferred embodiment, the step optimization model calculates the historical gradient information at iteration points. ,in , , , , These are the calculated coefficients obtained through prior training. In another preferred embodiment, the step optimization model calculates the historical gradient information at iteration points. ,in , , , These are the calculated coefficients obtained through prior training.

[0118] Example A6: Calculate the optimization model for the number of iteration points based on the functional relationship between spectral efficiency and the number of iteration point historical gradient information steps, the functional relationship between energy efficiency and the number of iteration point historical gradient information steps, and the positive correlation between these relationships and the optimization model for the number of iteration point historical gradient information steps.

[0119] Specifically, the functional relationship between spectral efficiency and the number of iterations for historical gradient information is expressed as: The functional relationship between energy efficiency and the number of iterations for historical gradient information is expressed as follows: Based on the functional relationship between spectral efficiency and the number of iterations using historical gradient information, and the positive correlation between energy efficiency and the number of iterations using historical gradient information, the optimization model for the number of iterations using historical gradient information is calculated. In a preferred embodiment, the step optimization model calculates the historical gradient information at iteration points. ,in , , , , These are the calculated coefficients obtained through prior training. In another preferred embodiment, the step optimization model calculates the historical gradient information at iteration points. ,in , , , These are the calculated coefficients obtained through prior training.

[0120] Example A7: Calculate the optimization model for the number of iteration points based on the functional relationship between time delay and the number of iteration point historical gradient information steps, the functional relationship between spectral efficiency and the number of iteration point historical gradient information steps, the functional relationship between energy efficiency and the number of iteration point historical gradient information steps, and the positive correlation between these relationships and the optimization model for the number of iteration point historical gradient information steps.

[0121] Specifically, the functional relationship between time delay and the number of iterations for historical gradient information is expressed as: The functional relationship between spectral efficiency and the number of iterations for historical gradient information is expressed as follows: The functional relationship between energy efficiency and the number of iterations for historical gradient information is expressed as follows: Based on the functional relationships between time delay and the number of iteration point historical gradient information steps, the functional relationships between spectral efficiency and the number of iteration point historical gradient information steps, and the functional relationships between energy efficiency and the number of iteration point historical gradient information steps, and the positive correlation between these relationships and the optimization model for the number of iteration point historical gradient information steps, the optimization model for the number of iteration point historical gradient information steps is calculated. In a preferred embodiment, the step optimization model calculates the historical gradient information at iteration points. ,in , , , , , , These are the calculated coefficients obtained through prior training. In another preferred embodiment, the step optimization model calculates the historical gradient information at iteration points. ,in , , , , These are the calculated coefficients obtained through prior training.

[0122] The iteration point historical gradient information step optimization model is calculated according to the method described in any of the embodiments A1 to A7. .

[0123] In step S035, the step optimization model based on the historical gradient information of the iteration points is calculated. The maximum value, which corresponds to the number of steps of historical gradient information at the iteration point. This is the optimal gradient iteration step number, denoted as . .

[0124] In a preferred embodiment, step S04, calculating the optimal solution of the objective function of the transmitter's simulated precoding matrix based on the optimal gradient iteration step number, is illustrated in the flowchart below. Figure 6 As shown, the steps include:

[0125] Step 041: Construct the Riemannian complex circular manifold and initialize the iteration points;

[0126] Step 042: Calculate the cost function and Riemann gradient at the iteration point;

[0127] Step 043: Calculate the optimal step size for the current iteration point based on the optimal gradient iteration step number;

[0128] Step 044: Calculate the optimal search direction for the current iteration point based on the Riemann gradient and the optimal step size for the current iteration point;

[0129] Step 045: Using the Armijo criterion-based line search, search in the tangent space according to the optimal search direction of the current iteration point and shrink back to the manifold to obtain the new iteration point;

[0130] Step 046: Store the gradient information of the current iteration point and update the optimal step size of the current iteration point;

[0131] Step 047: Determine whether the current iteration point meets the termination condition. If yes, stop the iteration counting and obtain the optimal solution of the objective function of the transmitter's simulated precoding matrix; otherwise, return to step 042. The termination condition is that the norm of the gradient reaches a preset value and the optimal step size of the current iteration point is less than the preset minimum step size.

[0132] In this embodiment, a non-monotonic linear search algorithm (RBB algorithm) is used to solve the objective function for optimizing the transmitter's hybrid precoding matrix. The solution process is described in detail below. The specific steps.

[0133] In step S041, a Riemannian complex circular manifold is constructed. Initialize iteration coefficients initial point , The number in the bottom right corner indicates the number of iterations.

[0134] The simulated beamforming design based on the RBB method will simulate the pre-encoder. Vectorization Its feasible region constitutes a complex circular manifold. ,in This represents the total number of elements in the matrix. cut space Includes all that satisfy tangent vector The transfer of search direction between adjacent iteration points is achieved by vector translation operation. In determining Search direction at the location Then, the tangent space vector Mapping back form: During the solution process, the simulated beamforming matrix needs to satisfy the constant modulus constraint, meaning that each element in the obtained simulated beamforming matrix must be a unit modulus. All optimization steps of the algorithm are performed on a complex circular manifold. The process proceeds on top of the above, and the iterative process automatically satisfies the requirements. That is, constant modulus constraint, no additional constraint processing is required, the constant modulus constraint in the simulation domain is transformed into an unconstrained optimization problem.

[0135] In step S042, the cost function and Riemann gradient at the iteration point are calculated. The Riemann gradient is the mapping of the Euclidean gradient to the tangent space. The relationship between the total differential and the Euclidean gradient is as follows: , express right The differential, The Euclidean gradient is .according to and The relationship between the total differential and the Euclidean gradient can be rewritten as follows: .

[0136] Expanding the objective function yields:

[0137] (1)

[0138] Differentiate the objective function term by term based on its expanded expression:

[0139] (2)

[0140] according to achievable ;

[0141] Therefore, we get:

[0142] (3)

[0143] Substituting formula (3) into the derivative of the objective function, we get:

[0144]

[0145] in , , .

[0146] Therefore, the Euclidean gradient matrix of the objective function is:

[0147] (4)

[0148] Complex circular manifold The Riemann gradient on the tangent space is projected by the Euclidean gradient onto the tangent space. The Riemann gradient is obtained:

[0149] (5)

[0150] In step S043, the optimal gradient iteration steps are determined based on the functional relationship between the number of iterations based on the historical gradient information of the iteration points and the step size in the non-monotonic linear search algorithm, and the aforementioned implementation method. Calculate the optimal step size, denoted as .

[0151] Non-monotonic linear search algorithm through or The constraints determine the step size factor, where For the increment of the iteration point, For gradient difference, In both cases, There are two different ways to solve this problem:

[0152] (6)

[0153] To avoid the algorithm getting trapped in local optima due to a fixed step size, the non-monotonic linear search algorithm employs an alternating odd-even iteration strategy, i.e.:

[0154] (7)

[0155] In step S044, the Hessian inverse matrix in the quasi-Newton method is approximated in scalar form. Calculate the optimal search direction for the current iteration point. , For the sake of simplicity, we use Indicates the optimal step size , It is the Riemann gradient of the objective function.

[0156] In step S045, the search is performed in the tangent space along the optimal search direction of the current iteration point. A line search based on the Armijo criterion is used to search for the next point in the tangent space, and then the search is folded back into the manifold to obtain the new iteration point. Specifically, from... The search begins at the point in the tangent space, and the optimal search direction at the current iteration point is determined. If the next point is selected according to the optimal step size, then the next point is... .

[0157] If the current point does not satisfy the Armijo criterion or exceeds the maximum number of iterations, stop the tangent space search.

[0158] Points obtained using line search in tangent space In this process, it is necessary to map this point from tangent space back to the manifold. , is represented as:

[0159] (8)

[0160] In step S046, the gradient information of the current iteration point and the gradient information of the historical iteration points are transmitted as vectors. Specifically, the gradient information of the current iteration point and the gradient information of the historical iteration points are transmitted as vectors. Historical gradient information includes the Riemann gradient. Gradient norm and directional derivative Transmit to tangent space It also stores the gradient information of the current iteration point.

[0161] Determine if the optimal step size at the current iteration point satisfies the step size constraint. If so, update the optimal step size at the current iteration point. Otherwise, update the optimal step size of the current iteration point to... .

[0162] In step S047, it is determined that the norm of the gradient has reached a preset value. Furthermore, if the optimal step size at the current iteration point is less than the preset minimum step size, then stop the iteration count and obtain the optimal solution for the objective function of the transmitter's hybrid precoding matrix optimization; otherwise, the iteration count continues. Increase by 1, then return to step S042.

[0163] In a preferred embodiment, step S05, obtaining the transmitter hybrid precoding matrix based on the optimal solution of the objective function of the transmitter analog precoding matrix, is illustrated in the flowchart below. Figure 7 As shown, the steps include:

[0164] Step S051: Obtain the transmitter simulation precoding matrix based on the optimal solution of the objective function of the transmitter simulation precoding matrix;

[0165] Step S052: Calculate the transmitter digital precoding matrix based on the optimal digital precoding matrix and the analog precoding matrix;

[0166] Step S053: Calculate the transmitter hybrid precoding matrix based on the transmitter analog precoding matrix and the transmitter digital precoding matrix.

[0167] In this embodiment, the simulated precoding matrix is ​​obtained by optimizing the objective function of the transmitter's hybrid precoding matrix. According to the optimal digital precoding matrix and analog precoding matrix Calculate the digital precoding matrix According to the transmitter's simulated precoding matrix and transmitter digital precoding matrix Obtain the transmitter hybrid precoding matrix .

[0168] In a preferred embodiment, step S06, performing hybrid beamforming based on the transmitter's hybrid precoding matrix, is illustrated in the flowchart below. Figure 8 As shown, it includes:

[0169] Step S061: Adjust the amplitude and phase of the transmitted signal according to the digital precoding matrix of the transmitter;

[0170] Step S062: Construct a phase shifter based on the transmitter's analog precoding matrix and connect it to the radio frequency link;

[0171] Step S063: After amplitude and phase adjustment and phase shifting, the transmitter signal is transmitted through the radio frequency antenna to form a mixed beam.

[0172] In this embodiment, the radio frequency link of the UAV transmitter is denoted as . The number of antennas is denoted as In a hybrid beamforming system, the raw data stream is represented as According to the dimension Transmitter digital precoding matrix Adjust the transmission signal The amplitude and phase, according to the dimension are Transmitter analog precoding matrix Construct a phase shifter and connect it to the RF link; transmit the signal. After amplitude and phase adjustment and phase shifting, a hybrid beam is formed via an RF antenna with a dimension of [missing information]. signal .

[0173] The transmitted signal of hybrid beamforming passes through the channel The dimension obtained after transmission is Received signal .in The vector is an additive white Gaussian noise that follows a complex Gaussian distribution. An analog precoder is used at the receiving end. Perform simulation domain processing, and then... After transmission via each RF link, it passes through a digital pre-encoder. Perform digital field processing to obtain the final result. signal The analog pre-encoder and digital pre-encoder Objective function construction and solution methods and simulation precoding matrix and digital precoding matrix same.

[0174] The performance of the hybrid beamforming method provided by this invention is evaluated through data and simulation experiments. In this embodiment, when The following table shows the spectral efficiency and latency of the algorithm under different historical gradient information steps in the range of -35 to -5. The spectral efficiency and latency in the table are averaged over each signal-to-noise ratio.

[0175]

[0176] The optimal number of historical gradient information steps is obtained according to the method described in the above embodiments. In the simulation experiment, the number of transmitting antennas and receiving antennas... Data streams and the number of transceiver radio frequency links The channel has scattering clusters and Individual diameter, average cluster power Arrival and departure azimuth angles and All follow a Laplace distribution, with a uniform average angle distribution, and the angle spread is... Analog precoder The initial phase at The internal parameters follow a uniform distribution, and the system modulation method is QPSK modulation. The maximum step size for linear search parameters is... minimum step size Initial step size Termination threshold The maximum number of linear search iterations is 1000.

[0177] Figure 9 The graphs show the spectral efficiency as a function of signal-to-noise ratio for different beamforming algorithms. Regarding spectral efficiency, the method described in this invention is compared with the all-digital SVD beamforming method (optimal performance). The corresponding all-digital hybrid beamforming method is compared with the Orthogonal Matching Pursuit (OMP) method and the Conjugate Gradient (CG) method. The spectral efficiency achieved by the method described in this invention is much higher than that of the OMP method. This is because the OMP method selects the point with the highest residual correlation in each iteration, which makes the method prone to getting trapped in local optima. In contrast, the method of this invention adopts a non-strictly monotonically decreasing line search strategy on the Riemannian manifold, which helps to avoid getting trapped in local optima at the iteration points.

[0178] Figure 10 The curves show the time delay as a function of signal-to-noise ratio for different beamforming algorithms. In terms of time delay, compared with the conjugate gradient CG method, it can be seen that the method described in this invention reduces the running time by 75.3% compared to the CG method. This is because the objective function of this method uses a least-squares solution, resulting in fewer objective function calls than the CG method; the method described in this invention employs a non-strictly monotonically decreasing line search strategy on the Riemannian manifold, which can accelerate global convergence by reducing the number of invalid iterations, thereby reducing time delay; the method described in this invention uses the current iteration point and the previous... The dynamic step size is calculated using gradient information at each iteration point, eliminating the need to calculate high-order matrices and resulting in low computational complexity. Therefore, the method described in this invention has a significantly lower latency than the CG method.

[0179] According to another embodiment of the present invention, a computer-readable storage medium is provided that stores a computer program for electronic data interchange, wherein the computer program causes a computer to execute the nonlinear hybrid beamforming method in unmanned aerial vehicle communication as described in any of the above embodiments.

[0180] According to another embodiment of the present invention, a nonlinear hybrid beamforming system for unmanned aerial vehicle (UAV) communication is provided, the structural schematic diagram of which is shown below. Figure 11 As shown, it includes:

[0181] processor;

[0182] Memory;

[0183] And one or more programs, wherein the one or more programs are stored in a memory and configured to be executed by the signal processing unit, the programs causing the computer to perform the nonlinear hybrid beamforming method in UAV communication as described in any of the above embodiments.

[0184] The methods described above according to the invention can be implemented in hardware, firmware, or as software or computer code that can be stored in a recording medium (such as a CD-ROM, RAM, floppy disk, hard disk, or magneto-optical disk), or as computer code originally stored on a remote recording medium or a non-transitory machine-readable medium and subsequently stored on a local recording medium, downloaded via a network. Thus, the methods described herein can be stored as software processing on a recording medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware (such as an AuIC or FPGA). It is understood that the computer, processor, microprocessor controller, or programmable hardware includes storage components (e.g., RAM, ROM, flash memory, etc.) capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods described herein. Furthermore, when a general-purpose computer accesses the code used to implement the processes shown herein, the execution of the code transforms the general-purpose computer into a dedicated computer for performing the processes shown herein.

[0185] Of course, those skilled in the art should recognize that the above embodiments are only used to illustrate the present invention and are not intended to limit the present invention. Any changes or modifications to the above embodiments that are within the scope of the present invention will fall within the protection scope of the present invention.

Claims

1. A nonlinear hybrid beamforming method in UAV communication, characterized in that, The method comprises the following steps: Step S01, calculating the constraint condition of the transmit end hybrid precoding matrix according to the communication channel state of the unmanned aerial vehicle and the state of the radio frequency antenna; Step S02, constructing the target function of the transmit end analog precoding matrix according to the constraint condition of the transmit end hybrid precoding matrix and the requirement of the optimal precoder; Step S03, calculating the optimal gradient iteration step number according to the relationship between the historical gradient information and the time delay / spectral efficiency / energy efficiency; Step S04, calculating the optimal solution of the transmit end analog precoding matrix target function according to the optimal gradient iteration step number; Step S05, obtaining the transmit end hybrid precoding matrix according to the optimal solution of the transmit end analog precoding matrix target function; Step S06, performing hybrid beamforming according to the transmit end hybrid precoding matrix.

2. The nonlinear hybrid beamforming method in UAV communications according to claim 1, wherein, The constraint condition of the transmit end hybrid precoding matrix calculated according to the communication channel state of the unmanned aerial vehicle and the state of the radio frequency antenna comprises: According to the unmanned aerial vehicle communication channel state and the radio frequency antenna state, an unmanned aerial vehicle transmitting end power constraint constant is calculated, denoted as ; The power constraint condition of the transmitting end is calculated according to the Frobenius norm wherein, denotes the analog precoding matrix of the transmitting end, denotes the digital precoding matrix of the transmitting end, denotes the hybrid precoding matrix of the transmitting end; the hybrid precoding matrix of the transmitting end is obtained by multiplying the digital precoding matrix of the transmitting end and the analog precoding matrix of the transmitting end.

3. The method of claim 2, wherein, The target function of the transmit end analog precoding matrix constructed according to the constraint condition of the transmit end hybrid precoding matrix and the requirement of the optimal precoder comprises the following steps: According to singular value decomposition of a channel matrix, an optimal digital precoding matrix for full-digital beamforming is obtained ; According to the Euclidean distance minimization of the optimal digital precoding matrix and the transmit end hybrid precoding matrix, an initial objective function is constructed, denoted as: ; The initial objective function is The partial derivative is where is the Moore-Penrose pseudo-inverse of The objective function of the analog precoding matrix at the transmitting end is constructed according to the partial derivative solution of the initial objective function and the Frobenius norm, and is expressed as , , is the feasible solution set of the analog precoder with the constant modulus constraint.

4. The method of claim 1, wherein, The optimal gradient iteration step number calculated according to the relationship between the historical gradient information and the time delay / spectral efficiency / energy efficiency comprises the following steps: calculating the functional relationship between the time delay and the iteration point historical gradient information step number according to the relationship between the complexity of the non-monotonic linear search algorithm and the iteration point historical gradient information; calculating the functional relationship between the spectral efficiency and the iteration point historical gradient information step number according to the fitting calculation of the spectral efficiency corresponding to different iteration point historical gradient information step numbers; calculating the functional relationship between the energy efficiency and the iteration point historical gradient information step number according to the fitting calculation of the energy efficiency corresponding to different iteration point historical gradient information step numbers; constructing the iteration point historical gradient information step number optimization model according to the functional relationship between the time delay / spectral efficiency / energy efficiency and the iteration point historical gradient information step number; obtaining the optimal gradient iteration step number according to the iteration point historical gradient information step number corresponding to the maximum value of the iteration point historical gradient information step number optimization model.

5. The method of claim 4, wherein, The functional relationship between the time delay and the iteration point historical gradient information step number calculated according to the relationship between the complexity of the non-monotonic linear search algorithm and the iteration point historical gradient information comprises the following steps: calculating the inherent calculation complexity of the non-monotonic linear search algorithm; the inherent calculation complexity comprises the calculation complexity of the Euclidean gradient, the calculation complexity of the Riemann gradient, the calculation complexity of the line search, the calculation complexity of the orthogonal projection and the calculation complexity of the retraction; calculating the additional calculation complexity of the non-monotonic linear search algorithm according to the functional relationship between the historical gradient information step number and the step length; obtaining the complexity of the non-monotonic linear search algorithm according to the inherent calculation complexity of the non-monotonic linear search algorithm and the additional calculation complexity of the non-monotonic linear search algorithm; calculating the functional relationship between the time delay and the iteration point historical gradient information step number according to the influence of the non-monotonic linear search algorithm complexity on different iteration point historical gradient information step numbers.

6. The method of claim 1, wherein, The step S04, calculating the optimal solution of the transmit end analog precoding matrix target function according to the optimal gradient iteration step number, comprises: Step 041, constructing a Riemann complex circle manifold and initializing an iteration point; Step 042, calculating the cost function and the Riemann gradient of the iteration point; Step 043, calculating the optimal step size of the current iteration point according to the optimal gradient iteration step number; Step 044, calculating the optimal search direction of the current iteration point according to the Riemann gradient and the optimal step size of the current iteration point; Step 045, searching and retracting to the manifold in the tangent space according to the optimal search direction of the current iteration point by using the line search based on the Armijo criterion to obtain a new iteration point; Step 046, storing the gradient information of the current iteration point and updating the optimal step size of the current iteration point; Step 047, judging whether the current iteration point meets the termination condition, if yes, stopping iteration counting to obtain the optimal solution of the transmitter analog precoding matrix target function; otherwise, returning to step 042; the termination condition is that the norm of the gradient reaches a preset value and the optimal step size of the current iteration point is less than a preset minimum step size.

7. The method of claim 1, wherein, The transmitter hybrid precoding matrix obtained according to the optimal solution of the transmitter analog precoding matrix target function comprises the following steps: obtaining the transmitter analog precoding matrix according to the optimal solution of the transmitter analog precoding matrix target function; calculating the transmitter digital precoding matrix according to the optimal digital precoding matrix and the analog precoding matrix; calculating the transmitter hybrid precoding matrix according to the transmitter analog precoding matrix and the transmitter digital precoding matrix.

8. The method of claim 1, wherein, The mixed beamforming according to the transmitter hybrid precoding matrix comprises: adjusting the amplitude and phase of the transmitted signal according to the transmitter digital precoding matrix; constructing a phase shifter according to the transmitter analog precoding matrix and connecting the phase shifter with the radio frequency link; forming a mixed beam by the radio frequency antenna after the transmitter signal passes through the amplitude and phase adjustment and the phase shifter.

9. A computer readable storage medium storing a computer program for electronic data interchange, wherein, The computer program enables the computer to execute the method according to any one of claims 1-8. 10.A nonlinear hybrid beamforming system in UAV communications, characterized in that, comprise: a processor; a memory; and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the processor, and the programs enable the computer to execute the method according to any one of claims 1-8.

Images