CN112989701A - Wavelength tuning method of SGDBR tunable semiconductor laser - Google Patents
Wavelength tuning method of SGDBR tunable semiconductor laser Download PDFInfo
- Publication number
- CN112989701A CN112989701A CN202110275752.9A CN202110275752A CN112989701A CN 112989701 A CN112989701 A CN 112989701A CN 202110275752 A CN202110275752 A CN 202110275752A CN 112989701 A CN112989701 A CN 112989701A
- Authority
- CN
- China
- Prior art keywords
- wavelength
- sgdbr
- semiconductor laser
- tunable semiconductor
- laser
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- H—ELECTRICITY
- H01—ELECTRIC ELEMENTS
- H01S—DEVICES USING THE PROCESS OF LIGHT AMPLIFICATION BY STIMULATED EMISSION OF RADIATION [LASER] TO AMPLIFY OR GENERATE LIGHT; DEVICES USING STIMULATED EMISSION OF ELECTROMAGNETIC RADIATION IN WAVE RANGES OTHER THAN OPTICAL
- H01S5/00—Semiconductor lasers
- H01S5/06—Arrangements for controlling the laser output parameters, e.g. by operating on the active medium
-
- H—ELECTRICITY
- H01—ELECTRIC ELEMENTS
- H01S—DEVICES USING THE PROCESS OF LIGHT AMPLIFICATION BY STIMULATED EMISSION OF RADIATION [LASER] TO AMPLIFY OR GENERATE LIGHT; DEVICES USING STIMULATED EMISSION OF ELECTROMAGNETIC RADIATION IN WAVE RANGES OTHER THAN OPTICAL
- H01S5/00—Semiconductor lasers
- H01S5/10—Construction or shape of the optical resonator, e.g. extended or external cavity, coupled cavities, bent-guide, varying width, thickness or composition of the active region
- H01S5/14—External cavity lasers
- H01S5/141—External cavity lasers using a wavelength selective device, e.g. a grating or etalon
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2111/00—Details relating to CAD techniques
- G06F2111/06—Multi-objective optimisation, e.g. Pareto optimisation using simulated annealing [SA], ant colony algorithms or genetic algorithms [GA]
Abstract
The invention discloses a wavelength tuning method of an SGDBR tunable semiconductor laser, which belongs to the technical field of semiconductor photoelectron and comprises the following steps: directly testing a first current-wavelength corresponding relation of the laser to be tested; initializing a laser structure parameter, updating by using an optimization algorithm, minimizing an error between a current-wavelength corresponding relation obtained by executing the second fitting step and the first current-wavelength corresponding relation, and taking the structure parameter at the moment as a target structure parameter; the second fitting step is: respectively combining the laser structure parameters with the front and rear grating zone currents of each group, and then inputting the combined laser structure parameters into a wavelength prediction neural network to obtain corresponding output wavelengths, thereby obtaining a second current-wavelength corresponding relation; combining multiple groups of front and back grating zone currents with target structure parameters, respectively, and inputting the combined values into a wavelength prediction neural network to obtain an actual current-wavelength corresponding relation. The invention can quickly find the current-wavelength corresponding relation of the SGDBR laser and improve the wavelength prediction precision and efficiency.
Description
Technical Field
The invention belongs to the technical field of semiconductor photoelectron, and particularly relates to a wavelength tuning method of an SGDBR tunable semiconductor laser.
Background
The tunable semiconductor laser is one of key devices in a next-generation all-optical dynamic optical network, can be used for wavelength backup and inventory management, reduces the system cost, can provide functions of automatic wavelength configuration, tunable wavelength conversion, wavelength routing and the like in a reconfigurable optical network, and greatly increases the flexibility of the optical network. The tunable semiconductor laser with the monolithic integration structure has the advantages of wide tuning range, small volume, high wavelength tuning speed, easy combination with other devices and the like, and is a preferred scheme for photonic integration and dynamic optical network application.
SGDBR (tunable Sampled Distributed Bragg Reflector) tunable semiconductor lasers have received much attention and research because of their advantages of wide tuning range, small size, fast wavelength switching speed, and easy monolithic integration with other semiconductor devices. The basic structure of the SGDBR tunable semiconductor laser is shown in fig. 1, and includes a front grating region (FSG), an Active region (Active), a Phase region (Phase), and a rear grating Region (RSG) in sequence. A sampled grating is a special periodic grating structure formed by periodically removing regions in a uniform grating, the periodic modulation resulting in a comb-like reflection spectrum of the grating. Different sampling periods are selected in the front grating area and the rear grating area, the periods of the corresponding comb-shaped reflection spectrum sequences are staggered by a certain distance, and when a pair of spectrum peaks in the two comb-shaped reflection spectrum sequences coincide, a single output wavelength can be selected. When current is injected into the front grating region and the rear grating region, the effective refractive index of the passive waveguide region can be changed by utilizing the plasma effect of free carriers so as to control the position of a comb-shaped reflectance spectrum peak; the phase region is used for changing the cavity mode of the laser; different grating reflection peaks and cavity modes can be aligned by simultaneously changing the tuning currents of the front grating area, the back grating area and the phase area, and the tuning mode similar to the vernier effect can realize a larger wavelength tuning range under the condition of small injection current.
The corresponding relation between the output wavelength of the SGDBR tunable semiconductor laser and the injection currents of the front and rear grating regions is very complex and cannot be expressed by a basic algorithm. At present, researchers mainly design an automatic wavelength scanning control system of an SGDBR (glass distributed Bragg reflector) laser by using LabVIEW software, the system can automatically scan current of the laser, collect and analyze an output spectrum of the laser and generate a current-wavelength data query table of the laser, and the method is long in time consumption for finding the relation between the current and the wavelength by depending on experiments. The technical error during the actual manufacture of the SGDBR causes the change of the structural parameters of the laser, so even if the current-wavelength correspondence relationship of the same batch of SGDBR tunable semiconductor lasers is greatly different, therefore, a 'current-wavelength' data query table generated by one SGDBR tunable semiconductor laser is not suitable for another SGDBR tunable semiconductor laser.
Disclosure of Invention
Aiming at the defects and improvement requirements of the prior art, the invention provides a wavelength tuning method of an SGDBR tunable semiconductor laser, which aims to quickly find the current-wavelength corresponding relation of the SGDBR laser so as to improve the wavelength prediction precision and efficiency.
To achieve the above object, according to one aspect of the present invention, there is provided a wavelength tuning method of an SGDBR tunable semiconductor laser, comprising the steps of:
a first fitting step: initializing a plurality of groups of front and back grating region currents, and respectively testing corresponding output wavelengths of the laser to be tested under the front and back grating region currents, thereby obtaining a first current-wavelength corresponding relation;
a second fitting step: respectively combining the structural parameters of the laser to be tested with the currents of the front and rear grating regions in the first fitting step, and inputting the combined currents into a trained wavelength prediction neural network to obtain corresponding output wavelengths, thereby obtaining a second current-wavelength corresponding relation; the wavelength prediction neural network is used for predicting the output wavelength of the SGDBR tunable semiconductor laser by taking the structural parameters of the SGDBR tunable semiconductor laser and the currents of the front grating region and the rear grating region as input;
and (3) optimizing: initializing the structural parameters of the laser to be tested, updating the structural parameters of the laser to be tested by using a preset optimization algorithm, minimizing the error between the second current-wavelength corresponding relation and the first current-wavelength corresponding relation after the second fitting step is executed, and determining the structural parameters with the minimum error as target structural parameters;
a wavelength tuning step: and combining a plurality of groups of front and rear grating region currents with the target structure parameters respectively, and inputting the combined currents into a wavelength prediction neural network to obtain corresponding output wavelengths, thereby obtaining the actual current-wavelength corresponding relation of the laser to be measured.
In the invention, considering that besides the currents of the front and rear grating regions, the structural parameters also have influence on the output wavelength of the SGDBR tunable semiconductor laser, the prediction of the output wavelength is carried out by using the wavelength prediction neural network taking the currents of the front and rear grating regions and the structural parameters as input at the same time, the prediction precision can be ensured, and the prediction can be rapidly completed compared with the traditional method of generating a current-wavelength data query table by scanning analysis; on the basis, the current-wavelength corresponding relation of the SGDBR tunable semiconductor laser to be tested is actually measured, the laser structure parameter which enables the error between the current-wavelength corresponding relation obtained by using the wavelength prediction neural network and the actually measured current-wavelength corresponding relation to be minimum is determined by using the wavelength prediction neural network and the optimization algorithm to be used as the structure parameter of the SGDBR tunable semiconductor laser to be tested, the output wavelengths corresponding to a plurality of groups of front and rear grating region currents of the SGDBR tunable semiconductor laser to be tested are obtained again by using the wavelength prediction neural network in combination with the determined parameter, and the actual current-wavelength corresponding relation of the SGDBR tunable semiconductor laser to be tested can be accurately obtained; when the current-wavelength corresponding relation of the laser is actually measured, the accuracy of the accurately determined structural parameters can be ensured only by measuring a few groups (for example, dozens of groups) of currents and corresponding output wavelengths, and the wavelength prediction neural network can quickly complete wavelength prediction, so that the method can quickly determine the structural parameters of the laser to be measured and quickly find the actual current-wavelength corresponding relation of the laser to be measured; the invention considers the influence of the structural parameters of the laser on the output wavelength, therefore, the invention can effectively improve the prediction precision of the output wavelength.
Further, the method for training the wavelength prediction neural network comprises the following steps:
establishing an SGDBR simulation model for predicting the output wavelength of the SGDBR tunable semiconductor laser according to the structural parameters of the SGDBR tunable semiconductor laser and the current of the front and rear grating regions;
random structure parameters of a plurality of groups of SGDBR tunable semiconductor lasers and currents of front and rear grating regions, obtaining output wavelengths corresponding to all groups of parameters by using an SGDBR simulation model, taking each group of parameters as input, taking the corresponding output wavelengths as output, and constructing a training data set;
and training the neural network by utilizing the training data set, and obtaining the wavelength prediction neural network after the training is finished.
Ensuring the training effect of the wavelength prediction neural network, and needing a large amount of training data; according to the invention, multiple groups of structural parameters and output wavelengths corresponding to the currents of the front and rear grating regions are obtained through simulation by using the SGDBR simulation model as training data, and an actual SGDBR tunable semiconductor laser is not depended on, so that abundant training data can be ensured to be obtained, the obtaining time of the training data is effectively reduced, and the training effect and the training efficiency of a neural network are improved.
Further, when the SGDBR simulation model is established, the forward optical field intensity and the backward optical field intensity of the active region of the SGDBR tunable semiconductor laser are obtained by solving with a time domain traveling wave method, the reflection spectrum and the transmission spectrum of the front grating region of the SGDBR tunable semiconductor laser are obtained with a transmission matrix method, and the reflection spectrum and the transmission spectrum of the back grating region of the SGDBR tunable semiconductor laser are obtained with a transmission matrix method.
The traditional simulation model establishing method usually only adopts a time domain traveling wave method, but because the SGDBR tunable semiconductor laser is a sectional laser, the grating area adopts the traveling wave method to program complicatedly, and the calculation efficiency is not high; according to the invention, the frequency domain transmission characteristics of the front grating region and the rear grating region, namely the reflection spectrum and the transmission spectrum, are calculated by using a transmission matrix method, a complex coupling equation does not need to be solved, only a transmission matrix form of a scanning laser light field needs to be obtained, and the solution is more convenient; therefore, the invention solves the light field intensity of the active region by using a time domain analysis method, solves the frequency domain transmission characteristics of the front and rear grating regions by using a transmission matrix method, and effectively improves the establishing efficiency and the applicability of the simulation model of the SGDBR tunable semiconductor laser.
Further, when the output wavelength corresponding to each group of parameters is obtained by using an SGDBR simulation model, parallel calculation of matlab software accelerates simulation.
The invention completes the structure of the training data set by utilizing the parallel computing acceleration simulation function of matlab software, and can effectively accelerate the computing speed.
Further, the neural network is a cascaded neural network.
The invention uses the cascade neural network as the wavelength prediction neural network, and the cascade neural network has simple structure, so that the training process of the model can be simplified on the basis of ensuring the prediction precision.
Further, the preset optimization algorithm is a particle swarm algorithm.
The invention utilizes Particle Swarm Optimization (PSO) as an optimization algorithm in the process of determining the structural parameters of the laser, and can effectively improve the speed and the precision of the process.
Further, the optimizing step comprises:
(S1) taking the structural parameter of each laser to be measured as a particle, and initializing a plurality of particles in a preset range to form a particle swarm;
(S2) performing a second fitting step for each particle to obtain a corresponding second current-wavelength correspondence, thereby obtaining an optimal position of the particle group;
(S3) calculating an error between the second current-wavelength correspondence and the first current-wavelength correspondence corresponding to the particle at the optimal position, and if the calculated error is less than a preset threshold or reaches a maximum number of iterations, proceeding to step (S4); otherwise, after updating each particle, proceeding to step (S2) to start the next iteration;
(S4) determining the structural parameters corresponding to the particles at the optimal positions in the last iteration as target structural parameters;
and the nominal value of the structural parameter of the laser to be tested belongs to a preset range.
Further, the structural parameters are structural parameters of the front and back grating regions.
Because the tuning function of the SGDBR tunable semiconductor laser is specifically completed by the front grating region and the rear grating region, the structural parameters of the front grating region and the rear grating region which influence the output wavelength are mainly used; when the invention is used for predicting the wavelength of the wave, only the main influence factors in the structural parameters of the laser, namely the structural parameters of the front grating area and the rear grating area are considered, so that the prediction precision can be improved, and the influence on the prediction efficiency can be avoided.
Further, the structural parameters include: the coupling coefficient of the front grating area, the coupling coefficient of the back grating area, the duty ratio of the front grating area, the duty ratio of the back grating area, the period of the front grating area and the period of the back grating area.
The coupling coefficient of the front grating region, the coupling coefficient of the rear grating region, the duty ratio of the front grating region, the duty ratio of the rear grating region, the period of the front grating region and the period of the rear grating region are the most core structural parameters influencing the output wavelength of the SGDBR tunable semiconductor laser.
According to another aspect of the present invention, there is provided a computer-readable storage medium, characterized by comprising a stored computer program; when the computer program is executed by the processor, the computer program controls the device on which the computer readable storage medium is positioned to execute the wavelength tuning method of the SGDBR tunable semiconductor laser provided by the invention.
Generally, by the above technical solution conceived by the present invention, the following beneficial effects can be obtained:
(1) when the output wavelength of the SGDBR tunable semiconductor laser is predicted, the structural parameters of the SGDBR tunable semiconductor laser are considered, the prediction of the output wavelength is completed by specifically utilizing a neural network, and the accurate prediction of the output wavelength can be quickly completed; for a laser with an unknown structure, the current-wavelength corresponding relation of the laser is measured actually, the structural parameters of the laser are determined by means of a wavelength prediction neural network and an optimization algorithm, the complete current-wavelength corresponding relation of the laser can be found quickly by means of the wavelength prediction neural network in combination with the determined structural parameters, and the precision and the efficiency of wavelength prediction are improved effectively.
(2) According to the invention, multiple groups of structural parameters and output wavelengths corresponding to the currents of the front and rear grating regions are obtained by using the SGDBR simulation model as training data for training the neural network, so that a large amount of training data can be obtained without depending on an actual SGDBR tunable semiconductor laser, rich training data can be obtained, the acquisition time of the training data is effectively reduced, and the training effect and the training efficiency of the neural network are improved.
(3) According to the invention, the time domain analysis method is used for solving the light field intensity of the active region, the transmission matrix method is used for solving the frequency domain transmission characteristics of the front and rear grating regions, and the establishing efficiency and the applicability of the simulation model of the SGDBR tunable semiconductor laser are effectively improved.
Drawings
Fig. 1 is a schematic structural diagram of a conventional SGDBR tunable semiconductor laser;
fig. 2 is a flowchart of a wavelength tuning method of an SGDBR tunable semiconductor laser according to an embodiment of the present invention;
fig. 3 is a schematic diagram of a training data set obtained by using an SGDBR simulation model according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of a wavelength prediction neural network according to an embodiment of the present invention;
fig. 5 is a schematic diagram of an optimization procedure provided in the embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further 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. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
In the present application, the terms "first," "second," and the like (if any) in the description and the drawings are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order.
In order to solve the technical problems that the existing method for searching the current-wavelength corresponding relation of the SGDBR tunable semiconductor laser by depending on experiments consumes long time, and for a new SGDBR tunable semiconductor laser, a large amount of time is needed to rebuild the current-wavelength corresponding relation, the invention provides a wavelength tuning method of the SGDBR tunable semiconductor laser, which has the overall thought that: the neural network is used for taking the laser structure parameters, the front grating area current and the rear grating area current as input to predict the output wavelength, so that the influence of the laser structure parameters is considered when the output wavelength is predicted, and the wavelength prediction is completed quickly and accurately; for an SGDBR tunable semiconductor laser with an unknown structure, the current-wavelength corresponding relation of the SGDBR tunable semiconductor laser is measured actually, the structural parameters of the SGDBR tunable semiconductor laser are determined by means of a wavelength prediction neural network and an optimization algorithm, the complete current-wavelength corresponding relation of the SGDBR tunable semiconductor laser is quickly found by means of the wavelength prediction neural network in combination with the determined structural parameters, and the accuracy and the efficiency of wavelength prediction are effectively improved.
The following are examples.
Example 1:
a method for tuning the wavelength of an SGDBR tunable semiconductor laser, as shown in fig. 2, comprises the following steps:
a first fitting step: initializing a plurality of groups of front and back grating region currents, and respectively testing corresponding output wavelengths of the laser to be tested under the front and back grating region currents, thereby obtaining a first current-wavelength corresponding relation;
a second fitting step: respectively combining the structural parameters of the laser to be tested with the currents of the front and rear grating regions in the first fitting step, and inputting the combined currents into a trained wavelength prediction neural network to obtain corresponding output wavelengths, thereby obtaining a second current-wavelength corresponding relation; the wavelength prediction neural network is used for predicting the output wavelength of the SGDBR tunable semiconductor laser by taking the structural parameters of the SGDBR tunable semiconductor laser and the currents of the front grating region and the rear grating region as input;
and (3) optimizing: initializing the structural parameters of the laser to be tested, updating the structural parameters of the laser to be tested by using a preset optimization algorithm, minimizing the error between the second current-wavelength corresponding relation and the first current-wavelength corresponding relation after the second fitting step is executed, and determining the structural parameters with the minimum error as target structural parameters;
a wavelength tuning step: combining a plurality of groups of front and rear grating region currents with the target structure parameters respectively, and inputting the combined currents into the wavelength prediction neural network to obtain corresponding output wavelengths, so as to obtain the actual current-wavelength corresponding relation of the laser to be measured;
as an optional implementation manner, in the first fitting step of this embodiment, 20 sets of currents are specifically initialized, each set of current includes a front grating region current and a back grating region current, and the front grating region current and the back grating region current of the laser to be measured are set to be consistent with a certain set of currents, that is, the laser to be measured can output an output wavelength corresponding to the set of currents.
As an optional implementation manner, in this embodiment, the method for training the wavelength prediction neural network includes:
establishing an SGDBR simulation model for predicting the output wavelength of the SGDBR tunable semiconductor laser according to the structural parameters of the SGDBR tunable semiconductor laser and the current of the front and rear grating regions;
random structure parameters of a plurality of groups of SGDBR tunable semiconductor lasers and currents of front and rear grating regions, obtaining output wavelengths corresponding to all groups of parameters by using an SGDBR simulation model, taking each group of parameters as input, taking the corresponding output wavelengths as output, and constructing a training data set;
training a neural network by utilizing a training data set, and obtaining a wavelength prediction neural network after the training is finished;
in the embodiment, a plurality of groups of structural parameters and output wavelengths corresponding to the currents of the front and rear grating regions are obtained by simulation through an SGDBR simulation model and are used as training data for training a neural network, and an actual SGDBR tunable semiconductor laser is not relied on, so that rich training data can be obtained, the obtaining time of the training data can be effectively reduced, and the training effect and the training efficiency of the neural network can be improved;
considering that the SGDBR tunable semiconductor laser is a segmented laser, the grating region of which adopts a traveling wave method for programming is complex and the calculation efficiency is not high, as an optional implementation manner, in this embodiment, when the SGDBR simulation model is established, the forward optical field intensity and the backward optical field intensity of the active region of the SGDBR tunable semiconductor laser are obtained by solving with a time domain traveling wave method, the reflection spectrum and the transmission spectrum of the front grating region of the SGDBR tunable semiconductor laser are obtained by a transmission matrix method, and the reflection spectrum and the transmission spectrum of the rear grating region of the SGDBR tunable semiconductor laser are obtained by a transmission matrix method;
the active region adopts a time domain traveling wave method, which is based on solving a time domain coupling wave equation of an intracavity forward and backward transmission light field, and comprises the following specific steps:
wherein F (t, z) and R (t, z) are respectively a forward light field and a backward light field,
and
mean that the forward light field F (t, z) and the backward light field R (t, z) correspond to each otherThe spontaneous emission noise of (a) is,
is the group velocity (propagation velocity of the wave packet), ngIs the group index of refraction, delta is the detuning factor, G0Represents the net gain, κrfAnd kappafrRespectively representing the coupling coefficients of the front grating area and the rear grating area;
obtaining the forward light field intensity and the backward light field intensity of the active area by using a time domain traveling wave method, obtaining the transmission characteristics of the front grating area and the back grating area by using a transmission matrix method, namely a reflection spectrum and a transmission spectrum, converting the reflection spectrum and the transmission spectrum of the front grating area into a time domain, and performing convolution operation with the forward light field intensity of the active area to obtain a light field passing through the front grating area; converting the reflection spectrum and the transmission spectrum of the rear grating area into a time domain, and performing convolution operation on the reflection spectrum and the transmission spectrum and the intensity of a backward light field of the active area to obtain a light field passing through the rear grating area;
with the back and forth transmission of light in the laser cavity, the above process is iteratively executed until the laser is in a stable state, and at the moment, discrete Fourier transform is performed on the light field passing through the front grating to obtain the output spectrum of the laser, namely the output wavelength. In the embodiment, the frequency domain transmission characteristics, namely the reflection spectrum and the transmission spectrum, of the front grating region and the rear grating region are calculated by using a transmission matrix method, a complex coupling equation does not need to be solved, only a transmission matrix form of a scanning laser light field needs to be obtained, and the solution is convenient; therefore, in the embodiment, the time domain analysis method is used for solving the light field intensity of the active region, and the transmission matrix method is used for solving the frequency domain transmission characteristics of the front and rear grating regions, so that the establishing efficiency and the applicability of the simulation model of the SGDBR tunable semiconductor laser are effectively improved; it should be noted that this is only a preferred embodiment, and other simulation models that can predict the laser output wavelength according to the current combination of the front and back grating regions can also be used in the present invention.
In order to ensure the training effect of the model, in this embodiment, the constructed training data set collectively includes 20000 sets of parameters and corresponding output wavelengths, and in order to further increase the computation speed, when the SGDBR simulation model is used to obtain the output wavelengths corresponding to the sets of parameters, parallel computation of matlab software accelerates the simulation; the training data set for training the neural network is finally constructed as shown in fig. 3, wherein Ia and Ib represent the front grating region current and the back grating region current, respectively, and Lambda represents the output wavelength.
In order to simplify model training and facilitate subsequent reverse design for the current of the front and rear grating regions, optionally, in this embodiment, the wavelength prediction model is specifically a cascade neural network; after the training is finished, the established wavelength prediction model is specifically shown in fig. 4, wherein the wavelength prediction model comprises 5 hidden layers, each hidden layer comprises 15 neurons, 60 neurons, 160 neurons, 30 neurons and 10 neurons, a ReLu activation function layer and a batch normalization layer of Batchnormalization are added between each two layers of the neural network, the used optimizer is an Adam optimizer, and the loss function is a mean square error loss function; it should be noted that the description of the wavelength prediction model structure is only an alternative embodiment of the present invention, and in some other embodiments of the present invention, other neural networks may be used according to actual needs.
Optionally, in this embodiment, the optimization algorithm preset in the optimization step is a particle swarm algorithm, so as to effectively improve the speed and the accuracy of the process of determining the laser structure parameters;
correspondingly, the optimization steps are as shown in fig. 5, and specifically include:
(S1) taking the structural parameter of each laser to be measured as a particle, and initializing a plurality of particles in a preset range to form a particle swarm; the nominal value of the structural parameter of the laser to be tested belongs to the preset range; the error between the actual structural parameters of the laser and the nominal structural parameters is not too large, the initial values are set in the range of the nominal values of the structural parameters, and the actual structural parameters can be determined quickly in the optimization process;
(S2) performing a second fitting step for each particle to obtain a corresponding second current-wavelength correspondence, thereby obtaining an optimal position of the particle group;
(S3) calculating an error between the second current-wavelength correspondence and the first current-wavelength correspondence corresponding to the particle at the optimal position, and if the calculated error is less than a preset threshold or reaches a maximum number of iterations, proceeding to step (S4); otherwise, after updating each particle, proceeding to step (S2) to start the next iteration;
(S4) determining the structural parameters corresponding to the particles at the optimal positions in the last iteration as target structural parameters;
in this embodiment, when predicting the output wavelength of the SGDBR tunable semiconductor laser, the specifically considered structural parameters are structural parameters of the front and rear grating regions, which specifically include a coupling coefficient of the front grating region, a coupling coefficient of the rear grating region, a duty ratio of the front grating region, a duty ratio of the rear grating region, a period of the front grating region, and a period of the rear grating region;
in the embodiment, the particle swarm algorithm is used as an optimization algorithm in the reverse design process, and compared with a genetic algorithm, the particle swarm algorithm does not have cross (cross) and Mutation (Mutation) operations, so that the speed is higher and the precision is higher; it should be noted that in some other embodiments of the present invention, under the condition that the speed and the precision can meet the requirement, other optimization algorithms such as a genetic algorithm, a Direct Binary Search (DBS) and the like may also be used;
in this embodiment, the optimization step is specifically completed by using a particle swarm algorithm toolbox in Matlab, and the conventional parameters of the particle swarm algorithm are set as follows: the particle dimension is 6 (corresponding to the coupling coefficient of the front grating region, the coupling coefficient of the back grating region, the duty ratio of the front grating region, the duty ratio of the back grating region, the period of the front grating region and the period of the back grating region, respectively), the number Y of particles is 24, the maximum number of iterations is set to 3000, the learning factor c1 is c2 is 2, and the target precision is 2e-10 (that is, the mean square error between the current-wavelength correspondence obtained by the neural network prediction and the current-wavelength correspondence tested is lower than 2 e-10).
In general, in this embodiment, the structural parameters of the SGDBR tunable semiconductor laser are determined, and then the output wavelength of the laser is predicted by using the neural network according to the front and rear grating region currents and the structural parameters, so that the accuracy and efficiency of predicting the wavelength of the SGDBR tunable semiconductor laser can be effectively improved.
Example 2:
a computer-readable storage medium comprising a stored computer program; when executed by a processor, the computer program controls an apparatus on which a computer-readable storage medium is stored to perform the wavelength tuning method for the SGDBR tunable semiconductor laser provided in embodiment 1 above.
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (10)
1. A wavelength tuning method of an SGDBR tunable semiconductor laser is characterized by comprising the following steps:
a first fitting step: initializing a plurality of groups of front and back grating region currents, and respectively testing corresponding output wavelengths of the laser to be tested under the front and back grating region currents, thereby obtaining a first current-wavelength corresponding relation;
a second fitting step: combining the structural parameters of the laser to be tested with the currents of the front and rear grating regions in the first fitting step respectively, and inputting the combined currents into a trained wavelength prediction neural network to obtain corresponding output wavelengths, thereby obtaining a second current-wavelength corresponding relation; the wavelength prediction neural network is used for predicting the output wavelength of the SGDBR tunable semiconductor laser by taking the structural parameters of the SGDBR tunable semiconductor laser and the currents of the front grating region and the rear grating region as input;
and (3) optimizing: initializing the structural parameters of the laser to be tested, updating the structural parameters of the laser to be tested by using a preset optimization algorithm, so that after the second fitting step is executed, the error between the second current-wavelength corresponding relation and the first current-wavelength corresponding relation is minimized, and determining the structural parameters with the minimum error as target structural parameters;
a wavelength tuning step: and combining a plurality of groups of front and rear grating region currents with the target structure parameters respectively, and inputting the combined currents into the wavelength prediction neural network to obtain corresponding output wavelengths, thereby obtaining the actual current-wavelength corresponding relation of the laser to be measured.
2. A method of wavelength tuning of an SGDBR tunable semiconductor laser as claimed in claim 1 wherein the method of training the wavelength predictive neural network comprises:
establishing an SGDBR simulation model for predicting the output wavelength of the SGDBR tunable semiconductor laser according to the structural parameters of the SGDBR tunable semiconductor laser and the current of the front and rear grating regions;
random structure parameters of a plurality of groups of SGDBR tunable semiconductor lasers and currents of front and rear grating regions, obtaining output wavelengths corresponding to all groups of parameters by using the SGDBR simulation model, taking each group of parameters as input, and taking the corresponding output wavelengths as output to construct a training data set;
and training the neural network by using the training data set, and obtaining the wavelength prediction neural network after the training is finished.
3. The method as claimed in claim 2, wherein the SGDBR simulation model is constructed such that the forward and backward optical field intensities of the active region of the SGDBR tunable semiconductor laser are obtained by a time domain traveling wave method, the reflection spectrum and the transmission spectrum of the front grating region of the SGDBR tunable semiconductor laser are obtained by a transmission matrix method, and the reflection spectrum and the transmission spectrum of the back grating region of the SGDBR tunable semiconductor laser are obtained by a transmission matrix method.
4. The wavelength tuning method of an SGDBR tunable semiconductor laser according to claim 2, wherein when the SGDBR simulation model is used to obtain the output wavelengths corresponding to each set of parameters, parallel computation accelerated simulation by matlab software is performed.
5. A method of wavelength tuning of an SGDBR tunable semiconductor laser as claimed in claim 2 wherein the neural network is a cascaded neural network.
6. The wavelength tuning method of an SGDBR tunable semiconductor laser as claimed in claim 1, wherein the predetermined optimization algorithm is a particle swarm algorithm.
7. A method of wavelength tuning an SGDBR tunable semiconductor laser as claimed in claim 6 wherein the step of optimizing comprises:
(S1) taking the structural parameter of each laser to be tested as a particle, and initializing a plurality of particles in a preset range to form a particle swarm;
(S2) for each particle, performing the second fitting step separately to obtain a corresponding second current-wavelength correspondence, thereby obtaining an optimal position of the particle population;
(S3) calculating an error between a second current-wavelength correspondence corresponding to the particle at the optimal position and the first current-wavelength correspondence, and if the calculated error is less than a preset threshold or reaches a maximum number of iterations, proceeding to step (S4); otherwise, after updating each particle, proceeding to step (S2) to start the next iteration;
(S4) determining the structural parameter corresponding to the particle at the optimal position in the last iteration as the target structural parameter;
and the nominal value of the structural parameter of the laser to be tested belongs to the preset range.
8. A method of wavelength tuning as claimed in any one of claims 1-7 wherein the said configuration parameters are configuration parameters of the front and back grating regions.
9. A method of wavelength tuning of an SGDBR tunable semiconductor laser as claimed in claim 8 wherein the structural parameters include: the coupling coefficient of the front grating area, the coupling coefficient of the back grating area, the duty ratio of the front grating area, the duty ratio of the back grating area, the period of the front grating area and the period of the back grating area.
10. A computer-readable storage medium comprising a stored computer program; the computer program, when executed by a processor, controls an apparatus on which the computer readable storage medium is located to perform a method of wavelength tuning an SGDBR tunable semiconductor laser as claimed in any one of claims 1 to 9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110275752.9A CN112989701B (en) | 2021-03-15 | 2021-03-15 | Wavelength tuning method of SGDBR tunable semiconductor laser |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110275752.9A CN112989701B (en) | 2021-03-15 | 2021-03-15 | Wavelength tuning method of SGDBR tunable semiconductor laser |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112989701A true CN112989701A (en) | 2021-06-18 |
| CN112989701B CN112989701B (en) | 2022-03-18 |
Family
ID=76335248
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110275752.9A Active CN112989701B (en) | 2021-03-15 | 2021-03-15 | Wavelength tuning method of SGDBR tunable semiconductor laser |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112989701B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114268006A (en) * | 2021-12-21 | 2022-04-01 | 北京邮电大学 | Wavelength tunable passive mode-locking fiber laser |
| CN116990905A (en) * | 2023-09-26 | 2023-11-03 | 之江实验室 | Refractive index adjustable sub-wavelength grating coupler and design method thereof |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH0856040A (en) * | 1994-08-15 | 1996-02-27 | Nippon Telegr & Teleph Corp <Ntt> | Wavelength control equipment of multielectrode laser diode |
| US20070297714A1 (en) * | 2006-06-12 | 2007-12-27 | University Of Missouri Rolla | Neural network demodulation for an optical sensor |
| US20160258741A1 (en) * | 2013-12-19 | 2016-09-08 | Halliburton Energy Services, Inc. | Using tunable lasers in the design, manufacture, and implementation of integrated optical elements |
| JP2020079980A (en) * | 2018-11-12 | 2020-05-28 | 日本電信電話株式会社 | Optical partial differential operation device and optical neural network |
-
2021
- 2021-03-15 CN CN202110275752.9A patent/CN112989701B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH0856040A (en) * | 1994-08-15 | 1996-02-27 | Nippon Telegr & Teleph Corp <Ntt> | Wavelength control equipment of multielectrode laser diode |
| US20070297714A1 (en) * | 2006-06-12 | 2007-12-27 | University Of Missouri Rolla | Neural network demodulation for an optical sensor |
| US20160258741A1 (en) * | 2013-12-19 | 2016-09-08 | Halliburton Energy Services, Inc. | Using tunable lasers in the design, manufacture, and implementation of integrated optical elements |
| JP2020079980A (en) * | 2018-11-12 | 2020-05-28 | 日本電信電話株式会社 | Optical partial differential operation device and optical neural network |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114268006A (en) * | 2021-12-21 | 2022-04-01 | 北京邮电大学 | Wavelength tunable passive mode-locking fiber laser |
| CN114268006B (en) * | 2021-12-21 | 2023-09-15 | 北京邮电大学 | Wavelength-tunable passive mode-locked fiber laser |
| CN116990905A (en) * | 2023-09-26 | 2023-11-03 | 之江实验室 | Refractive index adjustable sub-wavelength grating coupler and design method thereof |
| CN116990905B (en) * | 2023-09-26 | 2024-01-16 | 之江实验室 | Refractive index adjustable sub-wavelength grating coupler and design method thereof |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112989701B (en) | 2022-03-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112989701B (en) | Wavelength tuning method of SGDBR tunable semiconductor laser | |
| Vandoorne et al. | Parallel reservoir computing using optical amplifiers | |
| US11831505B2 (en) | Method and system of hybrid data-and-model-driven hierarchical network reconfiguration | |
| CN111224399B (en) | Method and system for analyzing power grid uncertainty trend | |
| Mihana et al. | Memory effect on adaptive decision making with a chaotic semiconductor laser | |
| CN111898316A (en) | Construction method and application of super-surface structure design model | |
| CN115906303A (en) | Planar microwave filter design method and device based on machine learning | |
| Xu et al. | Compromise policy for multi-stage stochastic linear programming: Variance and bias reduction | |
| CN114066014A (en) | Time sequence prediction method for machine learning based on asynchronous parallel multi-arm slot machine | |
| CN116662412B (en) | Data mining method for big data of power grid distribution and utilization | |
| CN105913161A (en) | Method of acquiring maximum power point of photovoltaic system based on multi-objective optimization | |
| Jordao et al. | Stage-wise neural architecture search | |
| Cerulli et al. | The k-labeled spanning forest problem | |
| Jiang et al. | Exploring high-performance slow light grating waveguides by means of deep learning | |
| Erni et al. | Application of evolutionary optimization algorithms in computational optics | |
| CN108108554B (en) | Multi-material vehicle body assembly sequence planning and optimizing method | |
| CN111985153B (en) | Silicon-based photonic device simulation method and system based on neural network and genetic algorithm | |
| CN116128034A (en) | Optical fiber structure two-dimensional feature extraction and optical characteristic prediction method based on convolutional neural network | |
| Takahashi et al. | Optimization of a photonic crystal nanocavity using covariance matrix adaptation evolution strategy | |
| Huang et al. | Optimal prediction intervals of wind power generation based on FA-ELM | |
| CN115564466A (en) | Double-layer day-ahead electricity price prediction method based on calibration window integration and coupled market characteristics | |
| Ding et al. | Optimizing equivalent circuit model parameters of DFB lasers with RSM model and NSGA-II algorithm | |
| CN115544869A (en) | Multi-target optimization method, system, equipment and medium for high-power microwave source | |
| Sun et al. | Design of the ultra-compact silicon multimode waveguide bends with arbitrary width and radius based on the back propagation neural network | |
| Fiers et al. | Optical information processing: Advances in nanophotonic reservoir computing |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |