Micro-ring resonator cavity time delay-bandwidth product maximization method and system

Abstract

The application discloses a micro-ring resonator time delay bandwidth product maximization method and system, the method comprises the following steps: obtaining a micro-ring resonator model, obtaining system parameters of the micro-ring resonator according to the micro-ring resonator model; inputting the system parameters of the micro-ring resonator into a DRRs-A3C algorithm model and running the DRRs-A3C algorithm model; obtaining the maximum time delay bandwidth product of the micro-ring resonator and the optimization parameter value corresponding to the maximum time delay bandwidth product according to the DRRs-A3C algorithm model. The micro-ring resonator model is first taken as a simulation environment of the DRRs-A3C algorithm, the parameters of the micro-ring resonator are optimized to obtain the maximum time delay bandwidth product and the corresponding coupling parameter value, the maximum time delay bandwidth product can improve the ability of the micro-ring resonator in optical signal processing, the parameter optimization result can be quickly converged through the DRRs-A3C algorithm model, and the convergence error is reduced.

Description

Technical Field

[0001] This invention relates to the field of micro-nano optics technology, and in particular to a method for maximizing the time-delay bandwidth product of a micro-ring resonator. Background Technology

[0002] Integrated photonics helps reduce footprint, weight, and power consumption while improving stability and performance. Microring resonators (MRRs), as important photonic components, offer advantages such as large delay, high Q value, high precision, ease of fabrication, and suitability for large-scale integration, making them ideal for reservoir network construction. A crucial parameter of MRRs is the Delay-Bandwidth Product (DBP), defined as follows: After an optical signal with a certain bandwidth passes through a MRR, the group delay at the resonant wavelength (i.e., the center frequency) is greater than that at non-resonant wavelengths, as observed in its group delay-frequency spectrum. As the wavelength moves further away from the center wavelength, the group delay gradually decreases. The frequency bandwidth corresponding to when the group delay is reduced to half its original value can be calculated. Multiplying this frequency bandwidth by the maximum group delay value at the center frequency gives the delay-bandwidth product. When MRRs are applied to fiber optic communication, given a fixed group delay, the signal bandwidth it can carry must be considered; therefore, obtaining the maximum delay-bandwidth product is of great significance. Existing solutions utilize the electromagnetically induced transparency (EIT) effect to calculate the time-delay bandwidth product of the resonant cavity. However, the time-delay bandwidth product calculated by the EIT effect is not the maximum time-delay bandwidth product of the micro-ring resonant cavity system. Alternatively, a genetic algorithm (GA) can be used to optimize the single micro-ring resonant cavity structure to calculate the maximum time-delay bandwidth product, but the GA optimization time is long and it is prone to obtaining local optima rather than global optima. Furthermore, dual-micro-ring resonant cavities and multi-micro-ring resonant cavities have larger time-delay bandwidth products, but the time-delay bandwidth product of dual-micro-ring resonant cavities has not been studied.

[0003] Therefore, existing technologies still need improvement. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method and system for maximizing the time delay bandwidth product of a microring resonator, in order to address the above-mentioned deficiencies of the prior art. The aim is to solve the problem that the maximum time delay bandwidth product of a microring resonator cannot be obtained quickly in the prior art.

[0005] The technical solution adopted by this invention to solve the problem is as follows:

[0006] First aspect: Embodiments of the present invention provide a method for maximizing the time-delay bandwidth product of a micro-ring resonator, wherein the method includes:

[0007] Obtain a micro-ring resonator model, and obtain the system parameters of the micro-ring resonator based on the micro-ring resonator model;

[0008] Input the system parameters of the micro-ring resonator into the DRRs-A3C algorithm model and run the DRRs-A3C algorithm model;

[0009] The maximum time delay bandwidth product of the micro-ring resonator and the optimized parameter values ​​corresponding to the maximum time delay bandwidth product are obtained based on the DRRs-A3C algorithm model.

[0010] In one implementation, obtaining the microring resonator model and obtaining the system parameters of the microring resonator based on the microring resonator model includes:

[0011] A microring resonant cavity model was established based on coupled-mode theory using finite element simulation software and numerical calculation software.

[0012] Run the microring resonator model to obtain the system parameters of the microring resonator model.

[0013] In one implementation, the operation of the DRRs-A3C algorithm model is as follows:

[0014] Initialize the global neural network parameters and worker neural network parameters of the DRRs-A3C algorithm model;

[0015] Input the state value S into the worker neural network of the DRRs-A3C algorithm model to obtain the action value A of the Actor neural network in the worker neural network. Input the action value A into the environment to obtain the reward R and the new state value S', and obtain the training set (S, A, R, S').

[0016] The state value S and the new state value S' in the training set are input into the Critic neural network in the worker neural network to obtain the evaluation value V(S) corresponding to the state value S and the new evaluation value V(S') corresponding to the new state value S', and the TD error δ=R+γ·V(S')-V(S) is calculated.

[0017] Iterate over the parameters in the worker neural network;

[0018] The iteration stops when the preset number of iterations is reached, and the global neural network parameters are updated.

[0019] Update the worker neural network parameters based on the global neural network parameters;

[0020] The worker neural network and the global neural network are trained iteratively and repeatedly.

[0021] When the global neural network converges, the DRRs-A3C algorithm model is output;

[0022] Wherein, δ is the TD error, R is the Critic neural network's estimate of the reward, and γ is the decay factor;

[0023] Both the global neural network and the worker neural network include an Actor network and a Critic network. The global neural network has the same structure as the worker neural network, and both the global neural network and the worker neural network are fully connected neural networks.

[0024] In one implementation, the environment in the DRRs-A3C algorithm model is a microring resonator system, the state value is the coupling coefficient of the microring resonator system, the action is the change in the coupling coefficient of the microring resonator system, and the reward is the time delay-bandwidth product and regularization term of the microring resonator.

[0025] In one implementation, obtaining the maximum time-delay bandwidth product of the micro-ring resonator and the optimized parameter value corresponding to the maximum time-delay bandwidth product based on the DRRs-A3C algorithm model includes:

[0026] The DRRs-A3C algorithm model parameters are initialized based on the system parameters of the micro-ring resonator.

[0027] The input state value is fed into the worker neural network of the DRRs-A3C algorithm model, and the worker neural network outputs a new state value.

[0028] The global neural network parameters of the DRRs-A3C algorithm model are updated according to the worker neural network parameters. The global neural network trains the new state values ​​and obtains the maximum time delay bandwidth product of the micro-ring resonator and the optimization parameters corresponding to the maximum time delay bandwidth product.

[0029] In one implementation, updating the global neural network parameters based on the worker neural network parameters, wherein the global neural network is trained to obtain the maximum time-delay bandwidth product and corresponding optimization parameters of the micro-ring resonator, includes:

[0030] Synchronize the parameters of the worker neural network to the global neural network;

[0031] The global neural network is trained to obtain the delay-bandwidth product and the corresponding optimization parameters;

[0032] Determine whether the global neural network has converged;

[0033] If the global neural network converges, it outputs the maximum time delay bandwidth product of the micro-ring resonator and the optimization parameters corresponding to the maximum time delay bandwidth product.

[0034] In one implementation, the micro-ring resonator is one or more of the following: a single-ring resonator, a cascaded double-ring resonator, a parallel double-ring resonator, a nested double-ring resonator, a 3×3 coupled double-ring resonator, or a multi-ring resonator composed of single and double-ring resonator topologies.

[0035] In one implementation, the DRRs-A3C algorithm model has 12 parallel workers.

[0036] In one implementation, the loss function of the Actor network in the DRRs-A3C algorithm model is loss = logπ. θ’ (S t ,A)·R, where the loss is the loss function of the Actor network, π θ’ The Actor network outputs the probabilities corresponding to the Gaussian probabilities in the action set. S represents all connection weight parameters of the Actor neural network in the Worker. t Let A be the state value at step t in the Worker, A be the change in the coupling coefficient in the Worker, and R be the reward estimate from the Critic neural network.

[0037] In one implementation, the formula for updating the Actor network weight parameters in the DRRs-A3C algorithm model is as follows: Among them, the Here, α represents the fully connected weight parameters in the Actor neural network, π represents the learning rate parameter, and π represents the weight parameters. θ’ The Actor network outputs the probabilities corresponding to the Gaussian probabilities in the action set. S represents all connection weight parameters of the Actor neural network in the Worker. t Let be the state value at step t in the Worker, A be the change in the coupling coefficient in the Worker, R be the reward estimate from the Critic neural network, and δ be the TD error.

[0038] Secondly, embodiments of the present invention also provide a system for maximizing the time-delay bandwidth product of a micro-ring resonator, wherein the system includes:

[0039] The acquisition module is used to acquire the micro-ring resonator model and acquire the system parameters of the micro-ring resonator based on the micro-ring resonator model.

[0040] The training module is used to input the system parameters of the microring resonator into the DRRs-A3C algorithm model and run the DRRs-A3C algorithm model;

[0041] The output module is used to obtain the maximum time delay bandwidth product of the micro-ring resonator and the optimized parameter value corresponding to the maximum time delay bandwidth product based on the DRRs-A3C algorithm model.

[0042] Thirdly, embodiments of the present invention also provide a terminal device, characterized in that the terminal device includes a memory, a processor, and a microring resonator delay-bandwidth product maximization program stored in the memory and executable on the processor, wherein when the processor executes the microring resonator delay-bandwidth product maximization program, it implements the steps of the microring resonator delay-bandwidth product maximization method of any of the above schemes.

[0043] Fourthly, embodiments of the present invention also provide a computer-readable storage medium, characterized in that the computer-readable storage medium stores a program for maximizing the delay-bandwidth product of a microring resonator, wherein when the program for maximizing the delay-bandwidth product of a microring resonator is executed by a processor, the steps of the method for maximizing the delay-bandwidth product of a microring resonator as described in any of the above schemes are implemented.

[0044] The beneficial effects of this invention are as follows: Compared with the prior art, this invention provides a method for maximizing the time-delay bandwidth product of a microring resonator. First, this invention obtains a microring resonator model and then acquires the system parameters of the microring resonator based on the model. Next, the system parameters of the microring resonator are input into a DRRs-A3C algorithm model and the DRRs-A3C algorithm model is run. Finally, the maximum time-delay bandwidth product of the microring resonator and the corresponding optimized parameter values ​​are obtained based on the DRRs-A3C algorithm model. This invention maximizes the time-delay bandwidth product of the microring resonator through the DRRs-A3C algorithm model and obtains the corresponding coupling parameter values. Maximizing the time-delay bandwidth product improves the optical signal processing capabilities of the microring resonator. The DRRs-A3C algorithm model allows for rapid convergence to obtain the optimized parameter results while reducing convergence errors. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 This is a flowchart illustrating the method for maximizing the time-delay bandwidth product of a microring resonator provided in an embodiment of the present invention.

[0047] Figure 2This is the physical model of the micro-ring resonator provided in the embodiments of the present invention.

[0048] Figure 3 This is a diagram of the DRRs-A3C algorithm model provided in an embodiment of the present invention.

[0049] Figure 4 This is an algorithm flowchart of the method for maximizing the time-delay bandwidth product of a micro-ring resonator provided in an embodiment of the present invention.

[0050] Figure 5 This is the time-delay-frequency spectrum of a double-cascaded microring resonator optimized by the time-delay-bandwidth product maximization method provided in the embodiments of the present invention.

[0051] Figure 6 This is a comparison chart of the inference time of the maximum delay bandwidth product of different optimization algorithms provided in the embodiments of the present invention for the cascaded dual microring resonant cavity structure.

[0052] Figure 7 This is a block diagram of the principle of the micro-ring resonator system for maximizing the time-delay bandwidth product provided in the embodiment of the present invention.

[0053] Figure 8 This is a schematic diagram of the internal structure of a terminal device that maximizes the time-delay bandwidth product of a micro-ring resonant cavity, as provided in an embodiment of the present invention. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0055] This embodiment provides a method for maximizing the time-delay bandwidth product of a microring resonator. This method maximizes the time-delay bandwidth product of the microring resonator. In specific implementation, firstly, a microring resonator model is obtained, and its system parameters are acquired based on the model. Then, the system parameters are input into a DRRs-A3C algorithm model and the model is run. Finally, the maximum time-delay bandwidth product and the corresponding optimized parameter values ​​are obtained from the DRRs-A3C algorithm model. This invention, by obtaining the system parameters of the microring resonator and inputting them into the DRRs-A3C algorithm model for optimization, quickly obtains the maximum time-delay bandwidth product and the corresponding coupling parameter values. Obtaining the maximum time-delay bandwidth product improves the microring resonator's optical signal processing capabilities. The DRRs-A3C algorithm model allows for rapid convergence of parameter optimization results while reducing convergence errors.

[0056] For example, such as Figure 2 (a) shows a cascaded dual-microring resonator system. The goal is to obtain its maximum time-delay bandwidth product and the corresponding coupling coefficient. First, a simulation model of the cascaded dual-microring resonator system is established and run to obtain the system parameters, namely the coupling coefficient of the microring resonators, the time delay at the center wavelength, the full width at half maximum (FWHM) of the time delay envelope at the center wavelength, and the time-delay bandwidth product. Then, the obtained system parameters are input into the DRRs-A3C algorithm model for optimization of the time-delay bandwidth product. After 300 epochs, the maximum time-delay bandwidth product is reached, which is 1395 ps·GHz. The corresponding coupling coefficients are t1 = 0.902, t2 = 0.011, and t3 = 0.354. This invention, by using the DRRs-A3C algorithm model, can quickly and conveniently optimize the coupling coefficient of the microring resonator to obtain the maximum time-delay bandwidth product.

[0057] Exemplary methods

[0058] This embodiment provides a method for maximizing the time-delay bandwidth product of a micro-ring resonator, which can be applied to terminal devices. Specifically, as follows... Figure 1 As shown, the method includes:

[0059] Step S100: Obtain the micro-ring resonator model, and obtain the system parameters of the micro-ring resonator based on the micro-ring resonator model.

[0060] In this embodiment, the microring resonator refers to an optical device composed of a straight waveguide and a ring waveguide, such as... Figure 2 As shown, the microring resonator can achieve delay and energy storage of optical signals. The microring resonator utilizes optical waveguides to form an optical circuit, creating feedback and thus influencing the output of light at specific wavelengths. Compared to traditional optical structures, the microring resonator is simpler and more integrated. Microring devices are fabricated using silicon material on an insulating substrate, and the process is compatible with CMOS technology. Mass production can significantly reduce the cost of optoelectronic devices. The system parameters include the coupling coefficient of the microring resonator, the time delay at the center wavelength, the full width at half maximum (FWHM) of the time delay envelope at the center wavelength, and the time delay-bandwidth product. The time delay is the output time difference between the optical signal entering the microring resonator and the optical signal not entering the microring resonator. The FWHM refers to the frequency width (also known as bandwidth, FWHM) when the transmittance or time delay in the transmission-frequency spectrum or time delay-frequency spectrum decreases from its highest value (at the center frequency) to half of its highest value. The time delay-bandwidth product is the product of the time delay and the bandwidth; multiplying the FWHM by the maximum group time delay value at the center frequency gives the time delay-bandwidth product.

[0061] In one implementation, the microring resonator model is established using finite element simulation software and numerical calculation software based on coupled-mode theory, and the coupling coefficient of the microring resonator is initialized; then, the system parameters of the microring resonator model are obtained by running the model. For example... Figure 2 The image shows the physical model of the microring resonator. Figure 2 In the diagram, the straight line represents a straight waveguide, and the ring shape represents a microring resonator. Light enters through the straight waveguide, is coupled through the microring resonator, and is then output.

[0062] In one implementation, the micro-ring resonator includes a single-ring resonator and a cascaded double-ring resonator (such as...). Figure 2 (a) shown), parallel double-ring resonant cavity (as shown in the image) Figure 2 (b) shown), nested double-ring resonant cavity (as shown in...) Figure 2 (c) shown), 3×3 coupled dual-ring resonant cavity (as shown in) Figure 2 (d) or one or more of the multiple ring resonators composed of single or double ring resonator topologies.

[0063] Step S200: Input the system parameters of the micro-ring resonator into the DRRs-A3C algorithm model and run the DRRs-A3C algorithm model.

[0064] In this embodiment, the training process of the DRRs-A3C algorithm model is specifically as follows: Figure 3 As shown in Figure 4: First, initialize the global neural network parameters θ and worker neural network parameters θ' of the DRRs-A3C algorithm model; where the network parameters of the Actor in the global neural network are θ and the network parameters of the Critic network are θ'. v In the worker neural network, the Actor network parameter is θ', and the Critic network parameter is θ'. v The DRRs-A3C algorithm model is initialized. Then, the worker neural network interacts with the environment, specifically with the micro-ring resonator system. The environment inputs the state value S into the worker neural network of the DRRs-A3C algorithm model, obtaining the action value A of the Actor neural network within the worker neural network. In other words, the coupling coefficient of the micro-ring resonator is input into the worker neural network of the DRRs-A3C algorithm model, and the worker neural network feeds back the change in the coupling coefficient. The output of the Actor network consists of two parts: one is the average value μ(s) of the action. t ;θ'), another output is the standard deviation of the action σ(s t ;θ'), and then the change in the coupling coefficient follows a normal distribution N(μ(s) t ;θ'),σ(s tAction A is obtained by random sampling on θ'). Then, the action value A is input into the environment to obtain the reward R and the new state value S', resulting in a training set (S, A, R, S'). The change in coupling coefficient is then re-input into the micro-ring resonator to calculate the time delay-bandwidth product and the new coupling coefficient. The state value S and the new state value S' in the training set are input into the Critic neural network in the worker neural network to obtain the evaluation value V(S) corresponding to the state value S and the new evaluation value V(S') corresponding to the new state value S', and the TD error δ = R + γ·V(S') - V(S) is calculated. The loss function of the Actor neural network in the worker neural network is iterated repeatedly. When the iteration reaches a preset number, it stops and the global neural network parameters are updated, that is, the convergence condition (s) of the Actor-Critic network model is reached. t n =terminal[0,1] or t=t max That is, when the state value (coupling coefficient t) is 0 or 1 at the boundary of the optimization constraint, or when the number of iterations reaches the set maximum number t = t max When convergence stops, the gradients can be calculated. Based on the gradients calculated at each time step, the cumulative gradients dθ and dθ' are then summed to calculate the cumulative gradients. v Based on the calculated dθ and dθ v For global neural network parameters θ and θ v Update complete, evaluation of network weight update: θ v =θ v -β·dθ v Action network weight update: θ = θ - α·dθ, where α and β are update coefficients. The worker neural network parameters are updated based on the global neural network parameters, i.e., θ′ = θ and θ v ′=θ v And set the cumulative gradient to 0, i.e., dθ←0 and dθ v←0, and then the worker neural network interacts with the environment; the worker neural network and the global neural network are trained iteratively; when the global neural network converges, the DRRs-A3C algorithm model is output. The experience playback method in this application adopts a strategy of combining cumulative experience and single-step experience. While pursuing the maximum global benefit, it also focuses on maximizing the DBP (fast convergence) of each iteration. Therefore, by adopting a strategy of combining cumulative experience and single-step experience, the number of global iterations can be set to 1000. Wherein, δ is the TD error, R is the Critic neural network's estimate of the reward, and γ is the decay factor; both the global neural network and the worker neural network include an Actor network and a Critic network. The global neural network has the same structure as the worker neural network, and both the global neural network and the worker neural network are fully connected neural networks. The parameter θ of the state value function of the global neural network. v The gradient descent method is used for updating, and the update formula is as follows: The policy π parameters of the global network are updated using the gradient ascent method. To overcome the algorithm's tendency to enter local optima and converge, the reward function is added to the objective function and directly bound to the policy π to improve exploration. The gradient of the entire objective function includes a regularization term related to the policy parameters, and the update formula is as follows: The loss function of the Critic network is calculated using the MSE method, and the weight parameters θ′ of the Critic network are updated. v The loss function for the Critic network is: loss = ∑(R + γV(S') - V(S)) 2 The formula for updating the weight parameters of the Critic network is:

[0065] In one implementation, the environment of the DRRs-A3C algorithm model is a microring resonator system, the state value is the coupling coefficient of the microring resonator system, the action is the change in the coupling coefficient of the microring resonator system, and the reward is the time-delay-bandwidth product and regularization term of the microring resonator. In this application, the environment is specifically set as a microring resonator system, and a model-based strategy is applied, enabling the DRRs-A3C algorithm in this application to converge quickly. Simultaneously, the coupling coefficient of the microring resonator is set as the state value, the change in the coupling coefficient of the microring resonator system is set as the action, and the time-delay-bandwidth product and regularization term of the microring resonator are set as the reward, i.e., R = DBP + λS.

[0066] In one implementation, the DRRs-A3C algorithm model uses 12 parallel workers. Specifically, this application sets the number of workers to 12. Since the number of workers affects the convergence error accuracy and convergence time of the algorithm, and there is a trade-off effect between the two, setting 12 workers in this application achieves optimal performance.

[0067] In one implementation, the loss function of the Actor network in the DRRs-A3C algorithm model is loss = logπ. θ’ (S t ,A)·R, where the loss is the loss function of the Actor network, π θ’ Let S be the probability corresponding to the Gaussian probability in the action set output by the Actor network, θ' be the weight parameters of all connections in the Actor neural network of the Worker, and S be the weight parameters of the Actor network of the Worker. t Let A be the state value at step t in the Worker, A be the change in the coupling coefficient in the Worker, and R be the reward estimate from the Critic neural network.

[0068] In one implementation, the formula for updating the Actor network weight parameters in the DRRs-A3C algorithm model is as follows: Among them, the Here, α represents the fully connected weight parameters in the Actor neural network, π represents the learning rate parameter, and π represents the weight parameters. θ’ The Actor network outputs the probabilities corresponding to the Gaussian probabilities in the action set. S represents all connection weight parameters of the Actor neural network in the Worker. t Let be the state value at step t in the Worker, A be the change in the coupling coefficient in the Worker, R be the reward estimate from the Critic neural network, and δ be the TD error.

[0069] Step S300: Obtain the maximum time delay bandwidth product of the micro-ring resonator and the optimized parameter value corresponding to the maximum time delay bandwidth product according to the DRRs-A3C algorithm model.

[0070] In this embodiment, the DRRs-A3C algorithm model parameters are initialized according to the system parameters of the micro-ring resonator; the state value is input to the worker neural network of the DRRs-A3C algorithm model, and the worker neural network outputs a new state value; the global neural network parameters of the DRRs-A3C algorithm model are updated according to the worker neural network parameters, the global neural network trains the new state value, and obtains the maximum time delay bandwidth product of the micro-ring resonator and the optimization parameters corresponding to the maximum time delay bandwidth product.

[0071] In one implementation, updating the global neural network parameters based on the worker neural network parameters, and training the global neural network to obtain the maximum time-delay bandwidth product and corresponding optimization parameters of the micro-ring resonator, includes: synchronizing the worker neural network parameters to the global neural network; training the global neural network to obtain the time-delay bandwidth product and the optimization parameters corresponding to the time-delay bandwidth product; determining whether the global neural network has converged; and if the global neural network has converged, outputting the maximum time-delay bandwidth product of the micro-ring resonator and the optimization parameters corresponding to the maximum time-delay bandwidth product.

[0072] In one implementation, the DRRs-A3C algorithm is used to optimize the parameters of the single microring resonator system. The physical structural parameters of the microring resonator are shown in Table 1. The microring resonator is modeled using a steady-state model (YARV model).

[0073] Table 1 Physical structural parameters of the microring resonator

[0074] waveguide materials Effective refractive index loss Resonant wavelength Micro-ring radius <![CDATA[LiNbO3]]> 1.9 1dB / cm 1550nm ≈70μm

[0075] First, the parameters of the single micro-ring resonator structure were optimized using the DRRs-A3C algorithm. After 100 epochs, the maximum delay-bandwidth product was achieved to be 605 ps·GHz, corresponding to a delay of 12.91 ps and a bandwidth of 49.9 GHz.

[0076] In another implementation, the DRRs-A3C algorithm is used to optimize the parameters of the cascaded dual-microring resonator system. After approximately 300 epochs, the maximum delay-bandwidth product is reached at 1395 ps·GHz, corresponding to a delay of 27.5 ps and a bandwidth of 50.6 GHz. The corresponding three coupling coefficients are t1 = 0.902, t2 = 0.011, and t3 = 0.354. Figure 5 The figure shows the time-delay-frequency spectrum of the dual-cascaded microring resonator after optimization using the DRRs-A3C algorithm. The thin black curve represents the time delay versus wavelength after random initialization, with a time-delay-bandwidth product of only 604 ps·GHz. The thick black curve represents the time delay after optimization using the DRRs-A3C algorithm. Figure 2 (a) After optimizing the delay-bandwidth product (DBP) of the structure, the group delay versus wavelength relationship is obtained, with the DBP reaching a maximum of 1395 ps·GHz. The cascaded dual-microring resonator structure readily achieves a "flat-top" result in the delay-frequency spectrum, resulting in a larger DBP, which is beneficial for constructing bandpass filters. This application still achieves a large DBP within different parameter ranges—that is, the robustness of the DBP to various coupling coefficients. For example... Figure 2(a) As shown in the diagram, the robustness of the cascaded dual-micro-ring structure is verified by using a delay-bandwidth product of 1380 ps·GHz as the minimum threshold. The value ranges of the three coupling coefficients are shown in Table 2.

[0077] Table 2. Robustness verification of coupling coefficients in the cascaded dual-micro-ring structure.

[0078] <![CDATA[t1]]> <![CDATA[t2]]> <![CDATA[t3]]> Delay-bandwidth product (ps·GHz) 0.898～0.910 0.011 0.354 >1380 0.902 0～0.03 0.354 >1380 0.902 0.011 0.335～0.370 >1380

[0079] In one implementation, the DRRs-A3C algorithm is used to optimize the parameters of the parallel dual-microring resonant cavity system. After approximately 300 epochs, the maximum delay-bandwidth product is reached, which is 1384 ps·GHz, corresponding to a delay of 24.8 ps and a bandwidth of 55.8 GHz. The corresponding four coupling coefficients are t1 = 0.8779, t2 = 0.8794, t3 = 0, and t4 = 0. For example... Figure 2 (b) As shown in the parallel dual-micro-ring structure, this invention uses a delay-bandwidth product of 1380 ps·GHz as the minimum threshold and verifies the robustness of the four coupling coefficients respectively. The value ranges of the four coupling coefficients are shown in Table 3.

[0080] Table 3. Robustness verification of coupling coefficients of parallel double microring structures.

[0081] <![CDATA[t1]]> <![CDATA[t2]]> <![CDATA[t3]]> <![CDATA[t4]]> Delay-bandwidth product (ps·GHz) 0.831～0.9 0.88 0 0 >1380 0.878 0.829～0.9 0 0 >1380 0.878 0.88 0～0.127 0 >1380 0.878 0.88 0 0～0.122 >1380

[0082] In one implementation, the DRRs-A3C algorithm is used to optimize the parameters of the nested dual-microring resonator system. After approximately 300 epochs, the maximum delay-bandwidth product is reached at 1390 ps·GHz, corresponding to a delay of 14.3 ps and a bandwidth of 97.1 GHz; the corresponding four coupling coefficients are t1 = 0.9906, t2 = 0.1449, t3 = 0, and t4 = 0.4929. For example... Figure 2 (c) As shown in the nested dual micro-ring structure, this invention uses a delay-bandwidth product of 1380 ps·GHz as the minimum threshold and verifies the robustness of the four coupling coefficients respectively. The value ranges of the four coupling coefficients are shown in Table 4.

[0083] Table 4. Robustness verification of coupling coefficients of nested double microring structures.

[0084] <![CDATA[t1]]> <![CDATA[t2]]> <![CDATA[t3]]> <![CDATA[t4]]> Delay-bandwidth product (ps·GHz) 0.989～0.992 0.1449 0 0.4929 >1380 0.9906 0.128～0.16 0 0.4929 >1380 0.9906 0.1449 0～0.15 0.4929 >1380 0.9906 0.1449 0 0.480～0.511 >1380

[0085] In one implementation, the DRRs-A3C algorithm is used to optimize the parameters of a 3×3 coupled dual-microring resonator system. After approximately 300 epochs, the maximum delay-bandwidth product is reached at 624 ps·GHz, corresponding to a delay of 18.9 ps and a bandwidth of 33.1 GHz. The corresponding three coupling coefficients are t1 = 0.675, t2 = 0.010, and t3 = 0.010. For example... Figure 2 (d) The 3×3 coupler-type dual micro-ring structure described in this invention uses a delay-bandwidth product of 620 ps·GHz as the minimum threshold. The robustness of the three coupling coefficients is verified separately, and the value ranges of the three coupling coefficients are shown in Table 5.

[0086] Table 5. Robustness verification of coupling coefficients of the 3×3 coupler-type dual microring structure.

[0087] <![CDATA[t1]]> <![CDATA[t2]]> <![CDATA[t3]]> Delay-bandwidth product (ps·GHz) 0.6～0.71 0.01 0.01 >620 0.675 0.006～0.017 0.01 >620 0.675 0.01 0.006～0.017 >620

[0088] In one implementation, the maximum delay-bandwidth product (DBP) and corresponding coupling coefficient values ​​of the dual-microring structure are shown in Table 6, optimized using the DRRs-A3C algorithm. The maximum DBP of the cascaded, parallel, and nested dual-microring resonator structures all exceed 1380 ps·GHz, indicating that the DBPs of these three structures are comparable. However, multiple optimization results consistently show that the cascaded dual-microring resonator structure is slightly larger than the other two structures. The maximum DBP of the 3×3 coupler-type dual-microring resonator structure is only 624 ps·GHz, comparable to the maximum DBP of a single-microring resonator.

[0089] Table 6. Maximum DBP parameter values ​​and optimal values ​​for four dual-micro-ring resonant cavity structures.

[0090]

[0091] In one implementation, Figure 6 The figure shows a comparison of the inference time (after optimization) of the maximum delay bandwidth product of different optimization algorithms for a cascaded dual-micro-ring resonant cavity structure. The Asynchronous Advantageous Actor-Critic Algorithm (DRRs-A3C) is faster in inference time than Simple Genetic Algorithm (SGA), Multi-Population Genetic Algorithm (MPGA), and Particle Swarm Optimization (PSO), requiring only about 5 minutes to complete the task, while the MPGA algorithm requires at least 20 minutes. At the same time, the convergence error of the DRRs-A3C algorithm is much smaller than that of PSO, SGA, and other algorithms, making it easier to find the global optimum.

[0092] Exemplary device

[0093] Based on the above embodiments, the present invention also provides a system for maximizing the time-delay-bandwidth product of a microring resonator, such as... Figure 7As shown, the system in this embodiment includes an acquisition module S10, a training module S20, and an output module S30. Specifically, the acquisition module is used to acquire a microring resonator model and obtain the system parameters of the microring resonator based on the microring resonator model. The training module is used to input the system parameters of the microring resonator into the DRRs-A3C algorithm model and train the DRRs-A3C algorithm model. The output module is used to obtain the maximum time-delay bandwidth product of the microring resonator and the optimized parameter values ​​corresponding to the maximum time-delay bandwidth product based on the DRRs-A3C algorithm model.

[0094] In one implementation, the system includes an acquisition module, the acquisition module comprising:

[0095] The model building unit is used to build a micro-ring resonant cavity model based on coupled-mode theory using finite element simulation software and numerical calculation software.

[0096] The model running unit is used to run the micro-ring resonator model and obtain the system parameters of the micro-ring resonator model.

[0097] In one implementation, the system includes a training module, the training module comprising:

[0098] Initialize the global neural network parameters and worker neural network parameters of the DRRs-A3C algorithm model;

[0099] Input the state value S into the worker neural network of the DRRs-A3C algorithm model to obtain the action value A of the Actor neural network in the worker neural network. Input the action value A into the environment to obtain the reward R and the new state value S', and obtain the training set (S, A, R, S').

[0100] The state value S and the new state value S' in the training set are input into the Critic neural network in the worker neural network to obtain the evaluation value V(S) corresponding to the state value S and the new evaluation value V(S') corresponding to the new state value S', and the TD error δ=R+γ·V(S')-V(S) is calculated.

[0101] Iterate over the parameters of the worker neural network;

[0102] The iteration stops when the preset number of iterations is reached, and the global neural network parameters are updated.

[0103] Update the worker neural network parameters based on the global neural network parameters;

[0104] The worker neural network and the global neural network are trained iteratively and repeatedly.

[0105] When the global neural network converges, the DRRs-A3C algorithm model is output;

[0106] Wherein, δ is the TD error, R is the Critic neural network's estimate of the reward, and γ is the decay factor;

[0107] Both the global neural network and the worker neural network include an Actor network and a Critic network. The global neural network has the same structure as the worker neural network, and both the global neural network and the worker neural network are fully connected neural networks.

[0108] In one implementation, the training module further includes:

[0109] In the DRRs-A3C algorithm model, the environment is a microring resonator system, the state value is the coupling coefficient of the microring resonator system, the action is the change in the coupling coefficient of the microring resonator system, and the reward is the time delay-bandwidth product and regularization term of the microring resonator.

[0110] In one implementation, the system includes an output module, the output module comprising:

[0111] An initialization unit is used to initialize the DRRs-A3C algorithm model parameters according to the system parameters of the micro-ring resonator.

[0112] A state output unit is used to input state values ​​into the worker neural network of the DRRs-A3C algorithm model, and the worker neural network outputs new state values.

[0113] The parameter update unit is used to update the global neural network parameters of the DRRs-A3C algorithm model according to the worker neural network parameters. The global neural network trains the new state value and obtains the maximum time delay bandwidth product of the micro-ring resonator and the optimization parameters corresponding to the maximum time delay bandwidth product.

[0114] In one implementation, the parameter update unit includes:

[0115] A parameter synchronization subunit is used to synchronize the parameters of the worker neural network to the global neural network;

[0116] The parameter optimization subunit is used to obtain the delay-bandwidth product and the optimized parameters corresponding to the delay-bandwidth product during the training of the global neural network.

[0117] A convergence determination subunit is used to determine whether the global neural network has converged;

[0118] The result output sub-unit is used to output the maximum time delay bandwidth product of the micro-ring resonator and the optimization parameters corresponding to the maximum time delay bandwidth product if the global neural network converges.

[0119] In one implementation, the system further includes:

[0120] The micro-ring resonator is one or more of the following: a single-ring resonator, a cascaded double-ring resonator, a parallel double-ring resonator, a nested double-ring resonator, a 3×3 coupled double-ring resonator, or a multi-ring resonator composed of single and double-ring resonator topologies.

[0121] In one implementation, the training module further includes:

[0122] The DRRs-A3C algorithm model has 12 parallel workers.

[0123] In one implementation, the training module further includes:

[0124] The loss function of the Actor network in the DRRs-A3C algorithm model is loss = logπ. θ’ (S t ,A)·R,

[0125] Wherein, the loss is the loss function of the Actor network, π θ 'θ' represents the probability corresponding to the Gaussian probability in the action set output by the Actor network, and θ' represents all connection weight parameters of the Actor neural network in the Worker. t Let A be the state value at step t in the Worker, A be the change in the coupling coefficient in the Worker, and R be the reward estimate from the Critic neural network.

[0126] In one implementation, the training module further includes:

[0127] The formula for updating the weight parameters of the Actor network in the DRRs-A3C algorithm model is as follows:

[0128] Among them, the Here, α represents the fully connected weight parameters in the Actor neural network, π represents the learning rate parameter, and π represents the weight parameters. θ’ S is the probability corresponding to the Gaussian probability in the action set output by the Actor network. t Let be the state value at step t in the Worker, A be the change in the coupling coefficient in the Worker, R be the reward estimate from the Critic neural network, and δ be the TD error.

[0129] The working principle of each module in the maintenance system of the terminal device in this embodiment is the same as that of each step in the above method embodiment, and will not be repeated here.

[0130] Based on the above embodiments, the present invention also provides a terminal device, the principle block diagram of which is shown in Figure 8. The terminal device may include one or more processors 100. Figure 8 (Only one is shown in the diagram), memory 101, and computer program 102 stored in memory 101 and executable on one or more processors 100. When one or more processors 100 execute computer program 102, they can implement the various steps in the embodiments. Alternatively, when one or more processors 100 execute computer program 102, they can implement the functions of various modules / units in the device embodiments, which is not limited here.

[0131] In one embodiment, the processor 100 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0132] In one embodiment, memory 101 may be an internal storage unit of an electronic device, such as a hard drive or RAM. Memory 101 may also be an external storage device of the electronic device, such as a plug-in hard drive, smart media card (SMC), secure digital (SD) card, flash card, etc. Furthermore, memory 101 may include both internal and external storage units. Memory 101 is used to store computer programs and other programs and data required by the terminal device. Memory 101 can also be used to temporarily store data that has been output or will be output.

[0133] Those skilled in the art will understand that Figure 8The block diagram shown is merely a partial structural diagram related to the present invention and does not constitute a limitation on the terminal device to which the present invention is applied. The specific terminal device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0134] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the methods described above. Any references to memory, storage, operational databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual operating data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0135] In summary, this invention discloses a method, system, terminal device, and storage medium for maximizing the delay-bandwidth product of a microring resonator. The method includes: acquiring a microring resonator model; acquiring system parameters of the microring resonator based on the model; inputting the system parameters into a DRRs-A3C algorithm model and running the model; and obtaining the maximum delay-bandwidth product and corresponding optimized parameter values ​​based on the DRRs-A3C algorithm model. This invention, after acquiring the system parameters of the microring resonator, inputs them into the DRRs-A3C algorithm model to quickly optimize the parameters, obtain the maximum delay-bandwidth product, and acquire the corresponding coupling parameter values. Maximizing the delay-bandwidth product improves the microring resonator's optical signal processing capabilities, and the DRRs-A3C algorithm model converges quickly to obtain the optimized parameter results while reducing convergence errors.

[0136] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for maximizing the time-delay bandwidth product of a microring resonator, characterized in that, The method includes: Obtain a micro-ring resonator model, and obtain the system parameters of the micro-ring resonator based on the micro-ring resonator model; Input the system parameters of the micro-ring resonator into the DRRs-A3C algorithm model and run the DRRs-A3C algorithm model; The specific operation process of the DRRs-A3C algorithm model is as follows: Initialize the global neural network parameters and worker neural network parameters of the DRRs-A3C algorithm model; Input the state value S into the worker neural network of the DRRs-A3C algorithm model to obtain the action value A of the Actor neural network in the worker neural network. Input the action value A into the environment to obtain the reward R and the new state value S', and obtain the training set (S, A, R, S'). The state value S and the new state value S' in the training set are input into the Critic neural network in the worker neural network to obtain the evaluation value V(S) corresponding to the state value S and the new evaluation value V(S') corresponding to the new state value S', and the TD error δ = R + γ·V(S') - V(S) is calculated. Iterate over the parameters of the worker neural network; The iteration stops when the preset number of iterations is reached, and the global neural network parameters are updated. Update the worker neural network parameters based on the global neural network parameters; The worker neural network and the global neural network are trained iteratively and repeatedly. When the global neural network converges, the DRRs-A3C algorithm model is output; Wherein, δ is the TD error, R is the Critic neural network's estimate of the reward, and γ is the decay factor; Both the global neural network and the worker neural network include an Actor network and a Critic network. The global neural network has the same structure as the worker neural network. Both the global neural network and the worker neural network are fully connected neural networks. The DRRs-A3C algorithm model parameters are initialized based on the system parameters of the micro-ring resonator. The input state value is fed into the worker neural network of the DRRs-A3C algorithm model, and the worker neural network outputs a new state value. The global neural network parameters of the DRRs-A3C algorithm model are updated according to the worker neural network parameters. The global neural network trains the new state values ​​and obtains the maximum time delay bandwidth product of the micro-ring resonator and the optimization parameters corresponding to the maximum time delay bandwidth product.

2. The method of claim 1, wherein the maximum time delay-bandwidth product of the micro- ring resonator is maximized by: The step of obtaining the micro-ring resonator model and obtaining the system parameters of the micro-ring resonator based on the micro-ring resonator model includes: A microring resonant cavity model was established based on coupled-mode theory using finite element simulation software and numerical calculation software. Run the microring resonator model to obtain the system parameters of the microring resonator model.

3. The method of claim 1, wherein the maximum time delay-bandwidth product of the micro- ring resonator is maximized by: In the DRRs-A3C algorithm model, the environment is a microring resonant cavity system, the state value is the coupling coefficient of the microring resonant cavity system, the action is the change in the coupling coefficient of the microring resonant cavity system, and the reward is the time delay-bandwidth product and regularization term of the microring resonant cavity.

4. The method of claim 1, wherein the maximum time delay-bandwidth product of the micro- ring resonator is maximized by: The step of updating the global neural network parameters based on the worker neural network parameters, and the global neural network training to obtain the maximum time delay bandwidth product and corresponding optimization parameters of the micro-ring resonator, includes: Synchronize the parameters of the worker neural network to the global neural network; The global neural network is trained to obtain the delay-bandwidth product and the corresponding optimization parameters; Determine whether the global neural network has converged; If the global neural network converges, it outputs the maximum time delay bandwidth product of the micro-ring resonator and the optimization parameters corresponding to the maximum time delay bandwidth product.

5. The method of claim 1, wherein the maximum time delay-bandwidth product of the micro- ring resonator is maximized by: The micro-ring resonator is one or more of the following: a single-ring resonator, a cascaded double-ring resonator, a parallel double-ring resonator, a nested double-ring resonator, a 3×3 coupled double-ring resonator, or a multi-ring resonator composed of single and double-ring resonator topologies.

6. The method for maximizing the time-delay bandwidth product of a microring resonator according to claim 1, characterized in that, The DRRs-A3C algorithm model has 12 parallel workers.

7. The method of claim 1, wherein the maximum time delay-bandwidth product of the micro- ring resonator is maximized by: The loss function of the Actor network in the DRRs-A3C algorithm model is loss = logπ θ’ (S t , A)·R, ​ Wherein, the loss is the loss function of the Actor network, π θ’ The Actor network outputs the probabilities corresponding to the Gaussian probabilities in the action set. θ ’ For all connection weight parameters of the Actor neural network in the Worker, S t Let A be the state value at step t in the Worker, A be the change in the coupling coefficient in the Worker, and R be the reward estimate from the Critic neural network.

8. The method of claim 1, wherein the maximum time delay-bandwidth product of the micro- ring resonator is maximized by: The Actor network weight parameter updating formula in the DRRs-A3C algorithm model is θ' = θ' +α · θ’ logπ θ’ (S t , A)·R·δ, Among them, the θ' These are the fully connected weight parameters in the Actor neural network. α π is the learning rate parameter. θ’ The Actor network outputs the probabilities corresponding to the Gaussian probabilities in the action set. S t Let be the state value at step t in the Worker, A be the change in the coupling coefficient in the Worker, R be the reward estimate from the Critic neural network, and δ be the TD error.

9. A micro-ring resonator cavity time-delay-bandwidth product maximization system, characterized by, The system includes: The acquisition module is used to acquire the micro-ring resonator model and acquire the system parameters of the micro-ring resonator based on the micro-ring resonator model. The training module is used to input the system parameters of the microring resonator into the DRRs-A3C algorithm model and run the DRRs-A3C algorithm model; The output module is used to obtain the maximum time delay bandwidth product of the micro-ring resonator and the optimized parameter value corresponding to the maximum time delay bandwidth product according to the DRRs-A3C algorithm model. The training module includes: Initialize the global neural network parameters and worker neural network parameters of the DRRs-A3C algorithm model; Input the state value S into the worker neural network of the DRRs-A3C algorithm model to obtain the action value A of the Actor neural network in the worker neural network. Input the action value A into the environment to obtain the reward R and the new state value S', and obtain the training set (S, A, R, S'). The state value S and the new state value S' in the training set are input into the Critic neural network in the worker neural network to obtain the evaluation value V(S) corresponding to the state value S and the new evaluation value V(S') corresponding to the new state value S', and the TD error δ = R + γ·V(S') - V(S) is calculated. Iterate over the parameters of the worker neural network; The iteration stops when the preset number of iterations is reached, and the global neural network parameters are updated. Update the worker neural network parameters based on the global neural network parameters; The worker neural network and the global neural network are trained iteratively and repeatedly. When the global neural network converges, the DRRs-A3C algorithm model is output; Wherein, δ is the TD error, R is the Critic neural network's estimate of the reward, and γ is the decay factor; Both the global neural network and the worker neural network include an Actor network and a Critic network. The global neural network has the same structure as the worker neural network. Both the global neural network and the worker neural network are fully connected neural networks. The output module includes: An initialization unit is used to initialize the DRRs-A3C algorithm model parameters according to the system parameters of the micro-ring resonator; A state output unit is used to input state values ​​into the worker neural network of the DRRs-A3C algorithm model, and the worker neural network outputs new state values. The parameter update unit is used to update the global neural network parameters of the DRRs-A3C algorithm model according to the worker neural network parameters. The global neural network trains the new state value and obtains the maximum time delay bandwidth product of the micro-ring resonator and the optimization parameters corresponding to the maximum time delay bandwidth product.

10. A terminal device, comprising: The terminal device includes a memory, a processor, and a microring resonator delay-bandwidth product maximization program stored in the memory and executable on the processor. When the processor executes the microring resonator delay-bandwidth product maximization program, it implements the steps of the microring resonator delay-bandwidth product maximization method as described in any one of claims 1-8.

11. A computer readable storage medium, characterized in that, The computer-readable storage medium stores a method program for maximizing the delay-bandwidth product of a microring resonator. When the method program is executed by a processor, it implements the steps of the method for maximizing the delay-bandwidth product of a microring resonator as described in any one of claims 1-8.

Images