Method and system for automatic finding of optimal efficiency point of a free space optical communication system

Abstract

This invention discloses an automatic method and system for finding the optimal efficiency point in a free-space optical communication system. The method includes hardware selection and initialization, with initialization encompassing system initialization and dataset initialization. It executes the current control command, reads the current power using an optical power meter, and adds the current power to the dataset. Bayesian optimization is used to find the next point to be searched. Fine-tuning is performed around the current maximum power point. The system includes a closed-loop control system and an optical power meter. This invention effectively avoids local optima and obtains the global maximum power point. Compared to a full traversal approach, it requires fewer samples and converges faster. In the fine-tuning stage, high-resolution optimization is achieved through local iterative updates, thus balancing speed and high accuracy.

Description

Technical Field

[0001] This invention relates to the field of optical pointing, capturing and tracking technology, and in particular to an automatic method and system for finding the optimal efficiency point in a free-space optical communication system. Background Technology

[0002] Because free-space optical links have small beam divergence angles and strong directivity, precise control of the pointing angle at the transmitter or receiver is usually required to obtain a stable high signal-to-noise ratio signal. In engineering implementation, two-dimensional actuators such as pan-tilt units, fast-reflecting mirrors, and MEMS mirrors are often used to adjust the two-dimensional angle of the beam or receiving field of view to achieve "pointing, acquisition, and tracking" (APT) of the optical link.

[0003] After initial acquisition, the system typically obtains a certain level of received optical power. At this point, how to automatically adjust the two-dimensional pointing to maximize the received power, given the existing optical power feedback, is a key issue in improving link margin and stability. In practical systems, received optical power, as the most readily obtainable one-dimensional feedback quantity, is widely used in closed-loop control.

[0004] Existing technologies involve scanning a two-dimensional angular space along a preset trajectory or grid after obtaining detectable optical power, measuring power at multiple directional positions, and finding an approximate maximum power point by comparing the power at different positions. These schemes typically employ a two-dimensional traversal or quasi-traversal scanning method combined with power comparison to achieve power optimization. This approach suffers from a large number of two-dimensional scan points, resulting in low power optimization efficiency. Furthermore, the algorithm only performs traversal or quasi-traversal searches, failing to fully utilize power feedback information. Moreover, it lacks robustness to noise and dynamic environments, making it difficult to guarantee global optimality. Summary of the Invention

[0005] Purpose of the invention: The purpose of this invention is to provide a method and system for automatically finding the optimal efficiency point in a free-space optical communication system.

[0006] Technical solution: The automatic method for finding the optimal efficiency point of a free-space optical communication system according to the present invention includes the following steps:

[0007] S1: Hardware selection and initialization, which includes system initialization and dataset initialization;

[0008] S2: Execute the current control command, read the current power through the optical power meter, and add the current power to the dataset;

[0009] S3: Use Bayesian optimization to find the next point to search;

[0010] S4: Perform fine-tuning around the current maximum power point.

[0011] Furthermore, step S1, system initialization, includes setting an initial pointing angle according to task requirements. That is, the initial horizontal angle of the platform. and vertical direction angle This initial pointing angle will serve as the starting point for subsequent adjustments.

[0012] Furthermore, step S1, dataset initialization, includes setting a preset initial pointing angle. Power measured by optical power meter To build.

[0013] Further, step S2 includes:

[0014] S21: Execute control commands;

[0015] S22: Measure the current optical power and update the dataset;

[0016] S23: Dwell time and ensuring stability;

[0017] S24: Determine whether the current power value meets the predetermined power requirement.

[0018] Furthermore, in step S21, the controller's task is to point the platform to the target angle and maintain that position for a period of time to obtain an error value.

[0019]

[0020] The updated pointing angle is:

[0021]

[0022]

[0023] in, It is the error between the target pointing angle and the current pointing angle; It is proportional gain. It is integral gain. It is the adjustment amount output by the controller; The updated pointing angle; The current pointing angle; The target pointing angle.

[0024] Further, in step S22, the optical power value at the current pointing angle is read by an optical power meter. , set the current pointing angle and the measured optical power value Add to the dataset to get the updated dataset. :

[0025]

[0026] The updated dataset provides new sample points for subsequent Bayesian optimization, which is used to optimize the selection of sampling points for the next iteration.

[0027] Furthermore, step S3, based on the existing optical power dataset, uses a Bayesian optimization algorithm to find the pointing angle that needs to be sampled next. Bayesian optimization models the "pointing angle-power" relationship using Gaussian process regression, predicts the possible maximum power point using known power feedback information, and selects the next sampling point by maximizing the acquisition function.

[0028]

[0029] Using the current pointing angle with the highest power as the center point:

[0030]

[0031] With the current optimal pointing angle Set the search boundaries in the horizontal and vertical directions, with the center as the center:

[0032]

[0033]

[0034] Form a two-dimensional pointing search region, according to a given step size , Generate grid points within the region:

[0035]

[0036]

[0037] in, Any candidate point is represented as:

[0038]

[0039] For each candidate point Calculate its predicted mean with standard deviation Substituting this into the desired improved formula, we get... ;

[0040]

[0041] in, The posterior mean is... The current known maximum power, To explore parameters, they are usually set to small values. For posterior variance, This is the cumulative distribution function.

[0042] Furthermore, step S4 uses the gradient ascent method to update the pointing angle to optimize the desired improvement function, determine the maximum power point, and the update formula is:

[0043]

[0044] in, and These are the gradients of the expected improvement function with respect to the horizontal and vertical directions, respectively.

[0045] In this way, the pointing angle is gradually adjusted to maximize the desired improvement value, thereby finding a more accurate maximum power point.

[0046] The automatic optimal efficiency point finding system for free-space optical communication systems described in this invention includes a closed-loop control system and an optical power meter. The closed-loop control system uses a pan-tilt unit, controller, dichroic mirror, lens, camera, and beacon light emitter to ensure that the optical platform can accurately point to a predetermined position. The optical power meter is used to measure the received optical power value of the platform at different pointing angles in real time, provide real-time feedback data, calculate the intensity of the optical signal, and provide power values ​​during the measurement process to support the optimization process.

[0047] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: The present invention is based on a strategy that combines Bayesian optimization and fine optimization. In the global search stage, it uses a surrogate model and a collection function to achieve adaptive exploration of the pointing space, which can effectively avoid local optima and obtain the global maximum power point. Compared with the full traversal method, it requires fewer sampling times and has a faster convergence speed. In the fine optimization stage, it achieves high-resolution optimization through local iterative updates, thus balancing speed and high accuracy. Attached Figure Description

[0048] Figure 1 This is a flowchart of the automatic method for finding the optimal efficiency point in the free-space optical communication system described in this invention.

[0049] Figure 2 This is a structural diagram of the automatic optimization point finding system for the free-space optical communication system described in this invention. Detailed Implementation

[0050] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0051] like Figure 2The diagram illustrates the structure of the automatic optimal efficiency point finding system for the free-space optical communication system described in this invention. It includes a closed-loop control system for precisely controlling the platform's pointing angle. This system automatically adjusts the platform's position based on the error between the current and target pointing angles, ensuring the platform stably points towards the target location. This control system can employ a pan-tilt unit, controller, dichroic mirror, lens, camera, beacon light emitter, or other mechanical devices to adjust the pointing angle, ensuring the optical platform accurately points to the predetermined position. An optical power meter is used to measure the received optical power value of the platform at different pointing angles in real time. The optical power meter provides real-time feedback data for calculating the intensity of the optical signal and provides necessary power values ​​during the measurement process to support optimization.

[0052] like Figure 1 As shown, the automatic method for finding the optimal efficiency point of a free-space optical communication system according to the present invention includes the following steps:

[0053] Step 1: Hardware selection and initialization, which includes system initialization and dataset initialization;

[0054] 1. System Initialization

[0055] When the system starts up, first set the target pointing angle according to the task requirements. That is, the initial horizontal angle of the platform. and vertical direction angle This initial pointing angle will serve as the starting point for subsequent adjustments.

[0056] In addition, to ensure the stable operation of the control system, the system performs necessary self-checks and calibrations, including: calibrating the pan-tilt unit or other actuators to ensure they can accurately adjust the platform's pointing angle under control commands; and calibrating the optical power meter to ensure it can accurately read optical power values ​​under different angles and environmental conditions.

[0057] 2. Dataset Initialization

[0058] To achieve effective Bayesian optimization, the dataset needs to be initialized. This dataset contains the platform pointing angle and corresponding optical power values. The initial dataset can be generated using a preset initial pointing angle. and power measured by an optical power meter To construct the simplest case, we can first choose an initial point:

[0059]

[0060] Point the gimbal Measure the corresponding optical power and record it as:

[0061]

[0062] Among more general methods, one can choose Initial points:

[0063]

[0064] In each Measure optical power at:

[0065]

[0066] This constitutes the initial dataset:

[0067]

[0068] Step 2: Execute the current control command, read the current power through the optical power meter, and add the current power to the dataset;

[0069] By executing control commands, the platform (such as a drone or optical equipment) controls the gimbal or other actuators based on the current pointing angle to track and stabilize the target position. An optical power meter measures the power at the target position and updates the dataset with the measured optical power value. The specific steps are as follows:

[0070] 1. Execute control commands (PI controller)

[0071] In step one, the platform's initial pointing angle The target angle has already been calculated, and in step two, the controller's task is to make the platform point accurately to this target angle and maintain that position for a period of time.

[0072] Obtain the error value:

[0073]

[0074] A PI (Proportional-Integral) controller is used to control the pointing angle of a platform, enabling it to quickly and stably reach the target pointing angle. The basic form of a PI controller is as follows:

[0075]

[0076] in: It is the error between the target pointing angle and the current pointing angle; The proportional gain determines the magnitude of the error in the response. It is the integral gain, which ensures the elimination of long-term system errors; It is the adjustment amount output by the controller, used to update the pointing angle of the platform.

[0077] The control process involves calculating errors. And adjust the pointing angle to drive the platform until the error... It is small enough that the platform stably points to the target location. The updated pointing angle is:

[0078]

[0079] The platform executes control commands to point to the current target location and maintains that location for a period of time to ensure the stability of the measurement.

[0080] 2. Measure the current optical power and update the dataset.

[0081] Once the platform is stably pointing at the target position, power is measured using an optical power meter. The optical power meter reads the optical power value at the current pointing angle. This is then used as the current measurement result. In practical applications, to ensure the accuracy of power measurement, the optical power is usually measured multiple times and the average value is taken. The calculation formula is as follows:

[0082]

[0083] in: For the first The power value measured this time; This refers to the number of power measurements performed. Multiple measurements can eliminate measurement noise and improve the stability and accuracy of the power data. [The current pointing angle is also specified.] and the measured power value Add to the dataset to get the updated dataset. :

[0084]

[0085] This dataset provides new sample points for subsequent Bayesian optimization, which can be used to optimize the selection of sampling points for the next iteration.

[0086] 3. Duration and stability assurance

[0087] When measuring power, the platform needs to remain stationary at the current target location for a certain period to ensure the optical power meter can read a stable power value. The duration of this stationary period is typically determined based on system stability and power fluctuations. Generally, the platform should remain stationary for at least [duration missing]. Seconds ensure the reliability of power measurement.

[0088] 4. Does it meet the power requirements?

[0089] After completing the power measurement, determine whether the current power value meets the predetermined power requirement. If the measured power To meet the needs, that is:

[0090]

[0091] Then proceed to step four (fine-tuning). If the power does not meet the requirements, proceed to step three (Bayesian optimization) for the next round of optimization.

[0092] Step 3: Use Bayesian optimization to find the next point to search;

[0093] Based on the existing optical power dataset, a Bayesian optimization algorithm is used to find the pointing angle for the next sampling iteration. Bayesian optimization models the pointing angle-power relationship using Gaussian process regression (GPR), predicts the possible maximum power point using known power feedback information, and selects the next sampling point by maximizing the acquisition function. The specific process is as follows:

[0094] Assumption:

[0095]

[0096] in: It is a mean function, and for simplicity, it is usually set to a constant or zero: , The covariance function (kernel function) characterizes the correlation of power values ​​between different pointing angles. In the two-dimensional pointing space, this invention preferably uses an anisotropic radial basis function (RBF) kernel, defined as:

[0097]

[0098] in: , This refers to the function value variance hyperparameter; These are the length-scale hyperparameters in the horizontal azimuth and vertical pitch directions, respectively. The marginal likelihood can be estimated offline or online by maximizing it. In practice, this invention preferably uses the method of maximizing the logarithmic marginal likelihood for automatic estimation.

[0099]

[0100] in: The sampled point angles are stacked row-wise into a matrix; For the corresponding power measurement vector; For the kernel matrix, the formulas for calculating the posterior mean and variance are:

[0101]

[0102]

[0103] in: The sampled pointing angles are stacked row-wise into a matrix: For the corresponding power measurement vector; The covariance matrix between the angle and the sampled points;

[0104] This is the covariance matrix of the sampled points.

[0105] The expected improvement (EI) function is shown below:

[0106]

[0107] in: This is the posterior mean (predicted power value). The current known maximum power, To explore parameters, they are usually set to small values ​​(e.g., 0.1). For the posterior variance (the uncertainty of power), This is the cumulative distribution function.

[0108] To obtain any point in two-dimensional space of After the expression, the position of the light spot to be pointed to next is determined in the following way: using the pointing angle with the highest current power as the center point:

[0109]

[0110]

[0111] For example, with the current optimal pointing angle Set the search boundaries in the horizontal and vertical directions, with the center as the center:

[0112]

[0113]

[0114] Form a two-dimensional pointing search area Select a set of candidate points within the search area:

[0115] Gridded candidate points: according to a given step size , Generate grid points within the region:

[0116]

[0117]

[0118] in, Any candidate point can be represented as:

[0119]

[0120] For each candidate point Calculate its predicted mean with standard deviation Then, substituting into the EI formula, we get... Finally, the point with the highest EI is selected:

[0121]

[0122] Once the next sampling point is determined In step two, the control system (such as a gimbal or actuator) adjusts the platform's pointing angle to... .

[0123] Step 4: Perform fine-tuning around the current maximum power point.

[0124] Gradient ascent optimizes the desired improvement (EI) function by updating the pointing angle to accurately determine the maximum power point. Assume the current pointing angle is... The updated formula is as follows:

[0125]

[0126] in: The current pointing angle, For learning rate, Let be the gradient of the desired improvement (EI) function. The desired improvement (EI) function is shown below:

[0127]

[0128] in: This is the posterior mean (predicted power value). The current known maximum power, To explore parameters, they are usually set to small values ​​(e.g., 0.1). For the posterior variance (the uncertainty of power), This is the cumulative distribution function.

[0129] Posterior mean and variance gradient:

[0130] In Gaussian process regression, the formulas for calculating the posterior mean and variance are:

[0131]

[0132]

[0133] in: This is the sampled pointing angle matrix. This is a power value vector. It is the covariance matrix between the pointing angle and the sampled points; This is the covariance matrix of the sampled points. Calculate... gradient

[0134]

[0135] The gradient can be calculated using the chain rule through the following steps:

[0136]

[0137]

[0138] in: The noise term is measured to control the effect of noise on the covariance. It is the identity matrix; : Power value vector, which records the power value at each sampling point.

[0139]

[0140]

[0141] Detailed update steps for the gradient ascent method:

[0142] Based on the gradient calculated above, the gradient ascent method can be used for updating. The update formula is:

[0143]

[0144] in, and These are the gradients of the expected improvement (EI) function relative to the horizontal and vertical directions, respectively. In this way, we gradually adjust the pointing angle. This maximizes the expected improvement (EI) value, thereby finding a more accurate maximum power point.

Claims

1. A method for automatically finding the optimal efficiency point in a free-space optical communication system, characterized in that, Includes the following steps: S1: Hardware selection and initialization, which includes system initialization and dataset initialization; S2: Execute the current control command, read the current power through the optical power meter, and add the current power to the dataset; S3: Use Bayesian optimization to find the next point to search; S4: Perform fine-tuning around the current maximum power point.

2. The method for automatically finding the optimal efficiency point in a free-space optical communication system according to claim 1, characterized in that, The system initialization step S1 includes setting the initial pointing angle according to task requirements. That is, the initial horizontal angle of the platform. and vertical direction angle This initial pointing angle will serve as the starting point for subsequent adjustments.

3. The method for automatically finding the optimal efficiency point in a free-space optical communication system according to claim 1, characterized in that, Step S1, dataset initialization, includes setting a preset initial pointing angle. Power measured by optical power meter To build.

4. The method for automatically finding the optimal efficiency point in a free-space optical communication system according to claim 1, characterized in that, Step S2 includes: S21: Execute control commands; S22: Measure the current optical power and update the dataset; S23: Dwell time and ensuring stability; S24: Determine whether the current power value meets the predetermined power requirement.

5. The method for automatically finding the optimal efficiency point in a free-space optical communication system according to claim 4, characterized in that, In step S21, the controller's task is to point the platform to the target angle and maintain that position for a period of time to obtain an error value. The updated pointing angle is: in, It is the error between the target pointing angle and the current pointing angle; It is proportional gain. It is integral gain. It is the adjustment amount output by the controller; The updated pointing angle; The current pointing angle; The target pointing angle.

6. The method for automatically finding the optimal efficiency point in a free-space optical communication system according to claim 4, characterized in that, In step S22, the optical power value at the current pointing angle is read using an optical power meter. , set the current pointing angle and the measured optical power value Add to the dataset to get the updated dataset. : The updated dataset provides new sample points for subsequent Bayesian optimization, which is used to optimize the selection of sampling points for the next iteration.

7. The method for automatically finding the optimal efficiency point in a free-space optical communication system according to claim 1, characterized in that, Step S3, based on the existing optical power dataset, uses a Bayesian optimization algorithm to find the pointing angle to be sampled next. Bayesian optimization models the "pointing angle-power" relationship using Gaussian process regression, predicts the possible maximum power point using known power feedback information, and selects the next sampling point by maximizing the acquisition function. Using the current pointing angle with the highest power as the center point: With the current optimal pointing angle Set the search boundaries in the horizontal and vertical directions, with the center as the center: Form a two-dimensional pointing search region, according to a given step size , Generate grid points within the region: in, Any candidate point is represented as: For each candidate point Calculate its predicted mean with standard deviation Substituting this into the desired improved formula, we get... ; in, The posterior mean is... The current known maximum power, To explore parameters, they are usually set to small values. For posterior variance, This is the cumulative distribution function.

8. The method for automatically finding the optimal efficiency point in a free-space optical communication system according to claim 1, characterized in that, Step S4 uses the gradient ascent method to update the pointing angle to optimize the desired improvement function and determine the maximum power point. The update formula is as follows: in, and These are the gradients of the expected improvement function with respect to the horizontal and vertical directions, respectively. In this way, the pointing angle is gradually adjusted to maximize the desired improvement value, thereby finding a more accurate maximum power point.

9. An automatic optimal efficiency point finding system for a free-space optical communication system, characterized in that, It includes a closed-loop control system and an optical power meter; the closed-loop control system uses a gimbal, controller, dichroic mirror, lens, camera, and beacon light emitter to ensure that the optical platform can be accurately pointed to a predetermined position; the optical power meter is used to measure the received optical power value of the platform at different pointing angles in real time, provide real-time feedback data, calculate the intensity of the optical signal, and provide power value during the measurement process to support the optimization process.

Images