Coherent Free-Space Optical Communication System Based on Improved Gray Wolf Algorithm

Abstract

This invention relates to the field of optical communication technology. To address the technical problems of slow convergence speed and susceptibility to local optima in existing gray wolf algorithms, a coherent free-space optical communication system based on an improved gray wolf algorithm is proposed. By introducing inertial weights and a nonlinear decay model to construct a velocity-guided position update equation, and combining it with an improved DLH strategy based on a dynamic elite reinforcement mechanism, the algorithm effectively balances its global exploration and local exploitation capabilities, solving the problem of traditional algorithms easily getting trapped in local optima when dealing with high-order Zernike aberrations. At the same time, a population size reduction NPSR strategy based on a nonlinear sigmoid function is designed to rapidly reduce the population size in the later stages of iteration through a nonlinear approach, significantly reducing the running time of the SLAO system.

Description

Technical Field

[0001] This invention relates to the field of optical communication technology, and in particular to a coherent free-space optical communication system based on an improved gray wolf algorithm. Background Technology

[0002] The surge in demand for ultra-high-speed, high-capacity connections in communication technologies has led to challenges such as spectrum congestion and electromagnetic interference for traditional wireless communications. Coherent Free Space Optical Communication (CFSOC) has attracted significant attention due to its superior transmission distance, sensitivity, and communication capacity compared to traditional free space optical communication. Compared to incoherent FSOC systems, which employ intensity detection / direct probes and binary on / off keying modulation, CFSOC systems offer numerous advantages, including flexible modulation / demodulation methods, high receiver sensitivity, and longer communication distances without repeaters.

[0003] This technology uses laser as the carrier wave and the atmosphere as the transmission medium. However, atmospheric turbulence distorts both the amplitude and wavefront of the laser carrier signal, leading to a decrease in system mixing efficiency (ME) and an increase in bit error rate (BER). Adaptive optics (AO) technology, by compensating for wavefront phase aberrations caused by atmospheric turbulence, is commonly used to improve CFSOC performance. Among these, sensorless adaptive optics (SLAO), a wavefront-free adaptive optics method, has received widespread attention in wavefront distortion correction for CFSOC systems because it does not require a wavefront sensor.

[0004] Existing sensorless adaptive optics SLAO techniques, such as model-based methods, are costly to implement; deep learning methods have weak generalization ability and poor correction effect under strong turbulence; while widely used model-free iterative algorithms (such as stochastic parallel gradient descent and its improved versions) face problems such as slow convergence speed, need for a large number of iterations and easy to get trapped in local optima, making it difficult to achieve efficient and robust wavefront distortion correction in strong atmospheric turbulence environments.

[0005] The Grey Wolf Algorithm (GWO) is a swarm intelligence optimization algorithm that simulates the hunting behavior of grey wolves and is used to solve various optimization problems. Traditional Grey Wolf Algorithms are limited by insufficient population diversity, resulting in slow convergence speed and the possibility of getting stuck in local optima. Summary of the Invention

[0006] This invention aims to solve the technical problems of slow convergence speed and easy getting trapped in local optima in existing algorithms, and proposes a coherent free space optical communication system based on the improved Grey Wolf algorithm.

[0007] A coherent free-space optical communication system based on the improved gray wolf algorithm includes a wavefront-free adaptive optics system for correcting wavefront distortion, wherein the wavefront-free adaptive optics system comprises:

[0008] A high-speed camera is used to acquire wavefront PSF images and transmit the preprocessed PSF images to the wavefront controller.

[0009] The wavefront controller is used to identify the wavefront features carried by the PSF image and generate a voltage signal to adjust the wavefront corrector to correct wavefront distortion based on the wavefront features.

[0010] The wavefront corrector receives the voltage signal and performs wavefront distortion correction, then outputs the corrected optical signal.

[0011] The wavefront controller internally runs an improved gray wolf algorithm, which combines velocity-guided updates, dimensionality-learning hunting strategies, and nonlinear population size reduction strategies on the basis of the gray wolf algorithm, thereby improving the efficiency and accuracy of wavefront distortion correction applications in wavefront sensorless adaptive optics systems.

[0012] Technical effects:

[0013] This invention combines an improved dimensionality learning-based hunting strategy with a nonlinear population size reduction technique to construct an improved Grey Wolf Algorithm (VDNGWO). It integrates three core strategies: velocity-guided update, dimensionality learning hunting (DLH), and nonlinear population size reduction (NPSR), and further optimizes algorithm performance through multi-strategy competitive selection and elite reinforcement mechanisms. By introducing inertia weights and a nonlinear decay model to construct a velocity-guided position update equation, and combining it with an improved DLH strategy based on a dynamic elite reinforcement mechanism, the algorithm effectively balances its global exploration and local exploitation capabilities, solving the problem of traditional algorithms easily getting trapped in local optima when dealing with high-order Zernike aberrations. Simultaneously, a population size reduction (NPSR) strategy based on a nonlinear sigmoid function is designed. Unlike traditional linear reduction, this strategy maintains sufficient population diversity in the early stages of the algorithm to ensure search breadth, and rapidly reduces the population size nonlinearly in the later stages of iteration, significantly reducing the computational overhead and runtime of the SLAO system.

[0014] Compared with existing mainstream wavefront-less adaptive optics control algorithms (such as SPGD and conventional GWO algorithms), this invention significantly improves convergence speed and computational efficiency. Existing technologies typically suffer from numerous iterations and high computational redundancy, making it difficult to meet the real-time requirements of high-speed communication. This invention, by introducing NPSR, maintains population diversity in the early stages of algorithm operation to ensure search breadth, while nonlinearly reducing the population size in the later stages to lower computational overhead. Simulation results show that, while achieving the same correction accuracy (e.g., RMS ≤ 0.1), the running time of this invention is significantly better than traditional algorithms, effectively solving the technical bottleneck of slow response speed in SLAO systems.

[0015] Furthermore, this invention significantly enhances the robustness and correction accuracy of the system under strong atmospheric turbulence. Addressing the weakness of existing algorithms that easily get trapped in local optima under strong turbulence, this invention constructs a velocity-guided position update mechanism and an improved dimensionality learning hunting (DLH) strategy, strengthening the algorithm's ability to escape local optima and achieving a dynamic balance between global exploration and local exploitation. Experiments demonstrate that even under strong turbulence conditions, this invention can stably improve the ME of the CFSOC system to above 0.99 and significantly reduce the BER, overcoming the instability of existing technologies under harsh atmospheric channels.

[0016] These improvements systematically address the issues of slow convergence and susceptibility to local optima in the traditional GWO algorithm when dealing with complex optimization problems from three aspects: position update mechanism, search strategy, and population management. They achieve a dynamic balance between global exploration and local exploitation, significantly improving the efficiency and accuracy of the algorithm in wavefront distortion correction applications in adaptive optics systems. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the structure of a CFSOC system with a SLAO system, which is a prior art technology.

[0018] Figure 2 This is a hierarchical diagram of the existing Grey Wolf algorithm.

[0019] Figure 3 This is a flowchart illustrating the operation of the VDNGWO algorithm in the wavefront controller in this embodiment.

[0020] Figure 4 The images show the initial Zernike coefficients for different wavefront aberrations under strong and weak turbulence, respectively.

[0021] Figure 5 The figures show the RMS variation with the number of iterations under strong and weak turbulence, respectively.

[0022] Figure 6 This is a comparison chart showing the effectiveness changes of different improved components in the VDNGWO algorithm.

[0023] Figure 7 RMS comparison plots of VDNGWO, GWO, SPGD and hybrid algorithms under strong turbulence and weak turbulence with different number of iterations.

[0024] Figure 8 This is a comparison of wavefront aberration phase diagrams and point spread functions under different turbulence intensities.

[0025] Figure 9 The wavefront aberration phase map and point spread function map are shown as a function of the number of iterations under strong turbulence conditions.

[0026] Figure 10 The wavefront aberration phase map and point spread function map are shown as the number of iterations under weak turbulence conditions. Detailed Implementation

[0027] The technical solution of the present invention will now be described in detail with reference to the accompanying drawings and preferred embodiments.

[0028] In this embodiment, a Continuous Surface Deformable Mirror (CSDM) is used as a wavefront corrector. The device consists of a two-dimensional array of voltage-driven actuators distributed on the optical pupil surface and coupled to a continuous reflective surface. By adjusting the driving voltage of each actuator in real time, the mirror shape can respond to changes in wavefront aberrations, thereby altering the optical path distribution on the pupil surface and achieving conjugate compensation for atmospheric turbulence aberrations. The mirror deformation of the CSDM can be expressed as a linear superposition of the influence functions of each actuator, where each influence function is theoretically approximated using a Gaussian mode. The specific calculation formula is as follows:

[0029] (1)

[0030] in This indicates the crosslinking value between adjacent actuators in the CSDM. This represents the normalized spacing between adjacent actuators. In CSDM, the first The coordinate position of each actuator These represent the Gaussian function coefficients. Under the driving voltage, the CSDM produces the following phase compensation for each actuator operation:

[0031] (2)

[0032] in Indicates the first Control voltage of each actuator This indicates the number of units in the CSDM actuator.

[0033] From a population perspective, gray wolves exhibit a pronounced gregarious lifestyle. Each standard wolf pack typically consists of 5-12 individuals, achieving a delicate balance between resource utilization efficiency and group management complexity. Of particular note is their social hierarchy, which gray wolf packs follow a strict social order. Based on different ranks, the pack can be divided into four levels: alpha wolf (leader), beta wolf (secondary leader), delta wolf (ordinary member), and omega wolf (lowest rank). The gray wolf pack hierarchy is as follows: Figure 2 As shown, this social structure and hunting strategy provide the mathematical model foundation for the GWO algorithm. Based on the social composition and behavioral patterns of gray wolf packs in nature, the hierarchy and hunting process of gray wolf packs are transformed into position update formulas through mathematical models.

[0034] During the prey-encircling phase, formulas (3) and (4) provide a mathematical model of the wolf pack's prey-encircling process:

[0035] (3)

[0036] (4)

[0037] in Indicates the current iteration number. Indicates the location of the prey. This indicates the location of the gray wolf. and The representative coefficient can be expressed as:

[0038] (5)

[0039] (6)

[0040] in The convergence factor is and The value is a random number within the interval [0,1]. The GWO algorithm considers the positions of the three wolves (α, β, δ) as optimal solutions during the iteration process. These three optimal solutions are saved, and ω wolf is forced to update its coordinates based on the position of the best search individual. The specific formula for this process is as follows:

[0041] (7)

[0042] (8)

[0043] (9)

[0044] This embodiment uses the VDNGWO algorithm, and its flowchart is as follows: Figure 3 As shown, the core strategy consists of three parts:

[0045] a. Position update equation based on velocity:

[0046] In the GWO algorithm, the convergence factor It is key to balancing exploration and development. When When the value is large, the population tends to explore the global solution space more comprehensively; while when When the value is small, it tends to focus on local development in the vicinity of the current region. In the traditional GWO algorithm, the convergence factor adopts a linear decreasing strategy, gradually decreasing from the initial value of 2 to 0, as expressed below:

[0047] (10)

[0048] in Indicates the maximum number of iterations. t This indicates the current iteration number.

[0049] A simple linear decreasing strategy cannot effectively adapt to the complex dynamic changes in the search process during actual optimization. This embodiment introduces inertia weights. And through nonlinear attenuation The model design update law coupled with inertia weights is shown in the following equation:

[0050] (11)

[0051] (12)

[0052] in It is a random number that takes a value in the interval [-1, 1]. It is a random number that takes a value in the interval [0,1]. This is the initial value for the weights, set to 0.9; is the final value of the weight, which is set to 0.1; e is the base of the natural logarithm, used to construct a non-linear curve and achieve a normalization effect.

[0053] By incorporating the global optimal solution during the speed update phase With two randomly selected population solutions and The velocity and position update formulas were reconstructed to improve local exploration efficiency. After optimization, the velocity update of an individual gray wolf is driven by the synergistic effect of inertial inheritance, population memory guidance, and individual memory interaction, as shown in the following expression:

[0054] (13)

[0055] (14)

[0056] In formula (13), It is the cognitive factor (set to 1.5). The social factor (set to 2.5). and It is a random number within the interval [0,1]. Velocity-inertia term. Inheriting the gray wolf speed information from the previous time step, among which express The speed. This represents the globally optimal solution for the current population. Population memory item. Guiding the gray wolf towards its historical optimal position, individual memory items This introduces inter-individual interactions within the population by utilizing the random difference between an individual's current position and its historical position.

[0057] b. Improved Dimensional Learning-Based Hunting Strategy (DLH):

[0058] DLH enhances the algorithm's exploration capabilities and maintains population diversity by improving the interaction between individual wolves and their neighbors and by introducing other randomly selected individuals from the pack to simulate individual hunting behavior.

[0059] In the DLH search strategy, new location The new location of an individual is calculated using formula (17), where the individual's new location is guided by its different neighbors and a wolf randomly selected from the population. During this process, the DLH search strategy also generates other reference locations to determine the wolf's new location, i.e. Therefore, the radius needs to be calculated. The radius passes through and The Euclidean distance between them is calculated as shown in formula (15):

[0060] (15)

[0061] Neighborhood set The result was then calculated using formula (16), which took into account... ,and Indicates that it is located at and The Euclidean distance between the wolves.

[0062] (16)

[0063] Subsequently, multi-neighborhood learning is performed using formula (17). Specifically, A random number within the interval [0,1]. The Weiyou Neighborhood The Maintenance and population The first wolf randomly selected from the middle The decision was made jointly by the two sides.

[0064] (17)

[0065] Finally, when selecting the optimal solution, the improved candidate solutions of the strategy need to be considered and evaluated by comparing their fitness values, as shown in Equation (18).

[0066] (18)

[0067] This embodiment employs a phased dynamic elite reinforcement mechanism based on the DLH strategy to further enhance the algorithm's adaptive optimization search capability and performance; this mechanism is referred to as the improved DLH. In the early search phase (t ≤ 25), individuals with poor fitness and performance in the population are reinitialized every three iterations. In the later iteration phase (t ≥ 40), candidate positions are finely adjusted by adding noise within 0.5% of the variance of the α wolf individuals, and fitness detection is combined to screen for better solutions.

[0068] c. Nonlinear population size reduction strategy (NPSR):

[0069] By using a nonlinear sigmoid function instead of a linear function as the benchmark, this method can be applied globally within the algorithm, and its mathematical expression is as follows:

[0070] (19)

[0071] in Indicates the population size for the next iteration; This represents the initial population size, set to 30. This represents the minimum population size required for convergence, which is set to 3. This represents the inflection point, i.e., the number of iterations at which the rate of population size decline reaches its maximum value, and is set to 20. The control parameter representing the rate of decline determines the steepness of the population size change, and its value is 0.4.

[0072] In the CFSOC system, ME and BER are key metrics for evaluating communication performance. By analyzing these key metrics, the optimal control effect of the VDNGWO algorithm in the SLAO system can be evaluated. Assuming the laser source is a plane wave with uniform beam intensity, according to coherent detection theory, the total optical power of the photodetector can be calculated using the following formula:

[0073] (20)

[0074] in and These represent the amplitudes of OS and LO, respectively. and These represent the frequencies of OS and LO, respectively. This represents the area of ​​the receiving aperture. It is the phase difference between OS and LO, where and These represent the phases of OS and LO, respectively. In the CFSOC system, phase distortion caused by atmospheric turbulence... It can be considered as a fixed state, and its calculation expression is as follows: (twenty one)

[0075] in This represents time-independent phase aberrations caused by atmospheric turbulence. This represents the modulation phase, which is independent of spatial coordinates. When = This is called zero-difference detection; when ≠ This is called heterodyne detection. Under zero-difference detection conditions, the measurement error of the CFSOC system is approximately equal to the Strel ratio (SR) of the far-field image, expressed by formula (22):

[0076] (twenty two)

[0077] SR is one of the parameters for measuring the quality of an optical system, defined as the ratio of the peak intensity of wavefront aberrations to the maximum intensity of ideal diffraction within the optical system. During propagation, the laser beam undergoes tilting and phase distortion due to atmospheric turbulence, thus increasing the BER. The expression for BER is given by the following equation:

[0078] (twenty three)

[0079] in Represents the complementary error function. This represents the signal-to-noise ratio under zero-difference detection conditions. Under atmospheric turbulence conditions... use The expression is as follows: (twenty four)

[0080] in This represents the number of photons received per bit. δ This indicates the quantum efficiency of the detector. This represents the optical power used to synchronize the bit error rate at the receiver in a binary phase-shift keying (BPSK) receiver system. Therefore, under atmospheric turbulence conditions, a BPSK receiver system can be expressed as: (25)

[0081] in η This represents the average error in the CFSOC system.

[0082] Zernike polynomials are commonly used to describe and fit wavefront aberrations caused by atmospheric turbulence in optical systems. These polynomials consist of a continuous sequence of mutually orthogonal polynomials defined on the unit circle, which can transform the distorted wavefront phase into a linear combination of weighted orthogonal polynomials, where each polynomial represents a type of aberration.

[0083] (26)

[0084] The constant term Indicates the piston term in the wavefront. and These represent tilt aberrations in the X and Y directions, respectively. These low-order aberrations can be directly corrected by BSU. Higher-order Zernike polynomials have higher fitting accuracy; however, wavefront aberration information affecting CFSOC communication performance is mainly concentrated in modes with lower Zernike aberrations. Therefore, 4th to 15th order Zernike polynomials are selected to model wavefront distortion caused by atmospheric turbulence.

[0085] parameter Used to quantify atmospheric turbulence intensity, by adjusting The values ​​can be used to obtain the Zenich polynomial coefficient distribution corresponding to different turbulence wavefront intensities. Atmospheric turbulence intensity is mainly divided into three categories: weak atmospheric turbulence corresponds to... The value is approximately 2, corresponding to moderate atmospheric turbulence. The value is approximately 10, corresponding to strong atmospheric turbulence. The value is greater than 15. The turbulence intensity can be expressed by combining equation (10) with the previous definition and by using a polynomial fitting method on the turbulent wavefront.

[0086] (27)

[0087] Here, σ φ Let RMS represent the root mean square of the residual wavefront phase obtained by Zernike fitting, and its expression is shown in Equation (28):

[0088] (28)

[0089] Based on this relationship, the root mean square (RMS) value of the system can be directly estimated using the Zernike coefficients. Using RMS as the core fitness value of the VDNGWO algorithm, the ME and BER of the CFSOC system are derived accordingly.

[0090] Will Set them to 20 and 5 respectively, with the default wavelength set to 635nm. Figure 4 As shown, Figure 4 In the table, (a) represents the Zernike polynomial coefficients generated under strong turbulence conditions, and (b) represents the Zernike polynomial coefficients generated under weak turbulence conditions. First, RMS was used as the algorithm's fitness index for evaluation. To ensure sufficient convergence, the maximum number of iterations was uniformly set to 50. Considering the randomness caused by noise and differences in initial values, 100 independent and repeated experiments were conducted for each turbulence condition. Figure 5 The results show the trend of the root mean square value as a function of the number of iterations under different turbulent conditions (i.e., strong turbulence (a) and weak turbulence (b)).

[0091] The results show that the VDNGWO algorithm can effectively converge within the preset number of iterations under both strong and weak turbulence conditions. In these 100 datasets, an average of 18 iterations are required to reduce the root mean square error (RMS) below 0.2, with a minimum of 11 iterations and a maximum of 24 iterations. Under strong turbulence conditions, an average of 26 iterations are required to reduce the RMS below 0.1, with a minimum of 18 iterations and a maximum of 39 iterations. Under weak turbulence conditions, an average of 6 iterations are required to reduce the RMS below 0.2 (minimum 3, maximum 8 iterations); while an average of 11 iterations are required to reduce it below 0.1 (minimum 7, maximum 15 iterations).

[0092] Independent numerical simulations were conducted for the three proposed improvement strategies to analyze the specific contribution of each strategy to the VDNGWO algorithm. All simulations used root mean square error as the optimization objective, and all other test conditions remained consistent with previous settings to ensure the comparability and reliability of the statistical results. The results for the first two improvement strategies are shown below. Figure 6 As shown, (a) is the position update equation based on velocity under strong turbulence; (b) is the DLH under strong turbulence; (c) is the improved DLH under strong turbulence; (d) is the position update equation based on velocity under weak turbulence; (e) is the DLH under weak turbulence; and (f) is the improved DLH under weak turbulence.

[0093] Figure 6 The velocity-based displacement update equation significantly accelerates the convergence process, but the algorithm is still prone to getting trapped in local optima. The original DLH strategy can partially avoid local optima, but the improvement is limited. When the improved DLH strategy is used alone, it is better than the original DLH in suppressing premature convergence. The improved algorithm based on these two strategies is called the improved GWO algorithm.

[0094] Building upon this, we further propose an optimization strategy based on LPSR—NPSR. To evaluate the efficiency improvement of NPSR compared to LPSR, we conduct tests under different conditions... Under these conditions, the performance of the algorithms was compared for various population reduction strategies. The comparison algorithms included: the enhanced GWO algorithm without population reduction, the LPSR-GWO algorithm with global LPSR application, and the MidLPSR-GWO algorithm with LPSR activated mid-iteration. All other test conditions remained consistent with previous results. Table 1 shows the average running time of each algorithm when the RMS is reduced to 0.1, and the average running time after 50 iterations. The running times of all algorithms were normalized to the running time of the improved GWO algorithm.

[0095] Table 1. Comparison of running times for different population size reduction strategies

[0096]

[0097] The results show that when At 20, the VDNGWO algorithm improved performance by 21.7% compared to the improved GWO algorithm, LPSR-GWO algorithm, and MidLPSR-GWO algorithm, and by 16.3% and 18.6% compared to the improved GWO algorithm, LPSR-GWO algorithm, and MidLPSR-GWO algorithm, respectively; when When the iteration count is 5, the running time of all algorithms is roughly equivalent, as they can reach the target metric within a relatively small number of iterations, making the population reduction strategy have minimal impact. When the number of iterations is fixed at 50, the VDNGWO algorithm maintains a smaller population size in the later stages, thereby reducing unnecessary computational overhead, and its running time is significantly lower than the other three algorithms—50.9%, 13.8%, and 37.7% shorter than the improved GWO, LPSR-GWO, and MidLPSR-GWO algorithms, respectively. It should be noted that although the LPSR-GWO algorithm also has an advantage in running time, its robustness is lower than other algorithms due to the reduced population diversity caused by the early reduction in population size; in contrast, the VDNGWO algorithm effectively avoids this problem while improving computational efficiency and maintaining good robustness.

[0098] To further evaluate the effectiveness of the VDNGWO algorithm, it was compared with existing SLAO control algorithms. Two representative algorithms successfully applied to wavefront aberration correction—the SPGD algorithm and a hybrid algorithm—were selected, with the unmodified GWO algorithm used as the benchmark. All algorithms were compared and simulated using RMS as the optimization objective. Referring to the parameter settings of existing studies, the gain coefficient λ of both the SPGD and hybrid algorithms was set to 2; the initial temperature of the hybrid algorithm was T=20°C, and the annealing efficiency γ=0.8. To ensure sufficient convergence space and facilitate the demonstration of the convergence process, the maximum number of iterations was set to 500, and each algorithm was run independently 30 times. The results are shown below. Figure 7 As shown, Figure 7 (a) shows the RMS curves of different iteration numbers under various control algorithms in strong turbulence conditions. Figure 7 (b) shows the RMS curves of different iterations for various control algorithms under weak turbulence conditions.

[0099] Simultaneously, using the previously generated Zernike coefficients, original wavefront aberration phase maps and corresponding point spread functions (PSFs) were constructed under different turbulence intensities, such as... Figure 8 As shown, (a) is the original phase diagram under strong turbulence; (b) is the original phase diagram under weak turbulence; (c) is the original point spread function under strong turbulence; and (d) is the original point spread function under weak turbulence. Without correction, the wavefront phase exhibits significant peaks and valleys, and the PSFs show a marked broadening of the focal plane intensity distribution and severe energy scattering, indicating significant wavefront aberrations in the system.

[0100] Figure 9 and Figure 10 The evolution of the residual wavefront aberration phase map and the corrected point spread function (PSF) during the iteration process is shown under different turbulence intensities, where (a)~(d) are phase maps and (e)~(h) are point spread functions. Under strong turbulence conditions, the residual aberration decreases significantly after 10 iterations; when the number of iterations increases to 20, the wavefront peak value further decreases, the PSF energy distribution shrinks significantly, and it gradually approaches ideal focusing; when the number of iterations increases to 50, the residual aberration tends to stabilize, indicating that the system performance is close to steady state. Under weak turbulence conditions, the algorithm converges faster and can obtain lower wavefront aberrations with fewer iterations.

[0101] Through the above simulation experiments, wavefront distortion was modeled using Zernike polynomials, and the correction effect of the VDNGWO algorithm was verified under different turbulence intensities. The algorithm was also compared with several traditional algorithms through theoretical analysis and numerical simulation. The results show that the VDNGWO algorithm maintains excellent convergence speed and robustness under strong atmospheric turbulence. This algorithm outperforms other algorithms in terms of the number of iterations and running time required to achieve the predetermined optimization targets. It maintains excellent performance even in complex and variable turbulent environments, effectively correcting wavefront distortion and improving the communication performance of the CFSOC system, thus enhancing the system's robustness.

[0102] It should be noted that other swarm intelligence optimization algorithms (such as Particle Swarm Optimization (PSO), Whale Optimization (WOA), Sparrow Search Algorithm (SSA), Harris Eagle Optimization (HHO), etc.) can also achieve similar wavefront correction effects after introducing the "velocity guidance mechanism" or "nonlinear population reduction strategy" proposed in this invention. All content not described in detail in this specification belongs to prior art known to those skilled in the art. Furthermore, for those skilled in the art, based on the ideas of this invention, there will be changes in specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A coherent free-space optical communication system based on an improved gray wolf algorithm, comprising a wavefront-free adaptive optics system for correcting wavefront distortion, wherein the wavefront-free adaptive optics system includes: A high-speed camera is used to acquire wavefront PSF images and transmit the preprocessed PSF images to the wavefront controller. The wavefront controller is used to identify the wavefront features carried by the PSF image and generate a voltage signal to adjust the wavefront corrector to correct wavefront distortion based on the wavefront features. The wavefront corrector receives the voltage signal and performs wavefront distortion correction, then outputs the corrected optical signal. The wavefront controller is characterized by internally running an improved gray wolf algorithm, which combines velocity-guided updates, dimensionality-learning hunting strategies, and nonlinear population size reduction strategies on the basis of the gray wolf algorithm structure. The improved Grey Wolf algorithm specifically includes the following speed-guided update: ， wherein, a is a non-linear decay, is a random number taking values in the interval [-1,1], is an inertia weight, t denotes the current iteration number, denotes the maximum number of iterations; By incorporating the global optimal solution in the velocity update phase with two randomly selected population solutions and , the velocity and position update formulas are reconstructed, and the specific expression formulas after optimization are as follows: ， Used to calculate the first step after the introduction of the speed guidance mechanism. i The final position of each individual gray wolf in the next iteration t+1; Among them, the velocity-inertia term Inheriting the gray wolf speed information from the previous time step, express Speed; population memory Guiding the gray wolf towards its historical optimal position, individual memory items By using the random difference between an individual's current position and its own historical position, inter-individual interactions are introduced within the population; It is a cognitive factor. It is a social factor; r 4 and r 5 is a random number within the interval [0,1]. The specific formula for the dimension-learning hunting strategy in the improved gray wolf algorithm is as follows: ， in, X i-GWO (t+1) represents the candidate position calculated by the traditional gray wolf optimization algorithm mechanism; X i-DLH (t+1) represents the candidate positions calculated based on the dimension-based learning hunting strategy; f() is the fitness function; The nonlinear population size reduction in the improved gray wolf algorithm specifically refers to: Using a nonlinear sigmoid function instead of a linear function as the benchmark, its mathematical expression is as follows: ， in Indicates the population size for the next iteration; This represents the initial population size, set to 30. This represents the minimum population size required for convergence, which is set to 3. This represents the inflection point, i.e., the number of iterations when the rate of population size decline reaches its maximum value, and is set to 20. The control parameter representing the rate of decline determines the steepness of the population size change, and its value is 0.

4.

2. The coherent free-space optical communication system based on the improved gray wolf algorithm according to claim 1, characterized in that, The inertial weight , in, It is a random number that takes a value in the interval [0,1]. This is the initial value for the weights, set to 0.9; This is the final value of the weight, taken as 0.1, where e is the base of the natural logarithm.

3. The coherent free-space optical communication system based on the improved gray wolf algorithm according to claim 1, characterized in that, In the dimensional learning hunting strategy, during the early search phase of the algorithm t ≤ 25, individuals with poor fitness and poor performance in the population are reinitialized every three iterations; during the later iteration phase of the algorithm t ≥ 40, candidate positions are finely adjusted by adding noise within 0.5% of the variance of α wolf individuals, and fitness detection is combined to screen for better solutions.

Images