Optical phased array phase calibration method and device, electronic device and storage medium
Through the target convolutional neural network model and residual module optimization feature extraction, the problem of insufficient accuracy of phase error calibration in optical phased arrays is solved, and efficient phase error calibration and rapid convergence are achieved.
Patent Information
- Application Number
- CN202510848129.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-24
- Publication Date
- 2025-07-22
- Estimated Expiration
- 2045-06-24
AI Technical Summary
The existing optical phase calibration algorithms fail to effectively solve the phase error problem introduced in large array arrays due to device process inhomogeneity, resulting in insufficient accuracy of phase error calibration.
The target convolutional neural network model is used to train the far-field training image and random phase perturbation image to establish a nonlinear mapping relationship between the far-field diffraction image and the phase, and the residual module and the sampling module are used to optimize feature extraction, combining grayscale processing and sampling segmentation to improve the accuracy of phase error calibration.
It effectively avoids the multi-solution problem of symmetric phase in optical phased arrays, improves the accuracy of phase error calibration, and achieves fast convergence and efficient phase calibration.
Smart Images

Figure CN120355605A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of optical detection, and particularly to an optical phased array phase calibration method, device, electronic device, and storage medium. Background Art
[0002] An optical phased array (OPA) is a technology based on the phased array principle that uses multiple optical transmitting units or optical receiving units to control the beam direction. It has advantages such as fast scanning speed, high robustness, and low sensitivity to environmental vibrations. However, in practical applications, to improve the resolution of the OPA, a large element array structure with more array elements is usually adopted to reduce the beam width. However, an OPA with a large element array, especially an array using a thermo-optic modulator, is prone to introduce phase error problems caused by device process non-uniformity.
[0003] Existing related calibration algorithms generally compensate for phase errors through a stochastic parallel gradient descent algorithm, but do not consider the problem that there may be non-unique solutions for phase errors in the optical phase space, resulting in the need to further improve the accuracy of phase error calibration of the optical phased array.
[0004] Regarding the problem that the accuracy of phase error calibration of the existing optical phased array in related technologies needs to be improved, no effective solution has been proposed yet. Summary of the Invention
[0005] In this embodiment, an optical phased array phase calibration method, device, electronic device, and storage medium are provided to solve the problem that the accuracy of phase error calibration of the existing optical phased array in related technologies needs to be improved.
[0006] In a first aspect, in this embodiment, an optical phased array phase calibration method is provided, which is applied to an optical phased array system. The optical phased array system includes an optical phased array and a target convolutional neural network model. The method includes:
[0007] Input a phase image to be calibrated into the optical phased array system to obtain a random perturbation phase image corresponding to the phase image to be calibrated. The phase image to be calibrated is generated according to the initial emission electric field in the optical phased array. The random perturbation phase image is obtained by performing phase perturbation within a preset range on the phase image to be calibrated.
[0008] Input the to-be-calibrated phase image and the random perturbation phase image into a target convolutional neural network model to determine the target phase error of the to-be-calibrated phase; the target convolutional neural network model is trained based on a pair of far-field training images for a preset convolutional neural network model; the pair of far-field training images includes a far-field training image generated according to an optical phased array and a far-field random perturbation phase image;
[0009] Calibrate the initial emission electric field of the to-be-calibrated phase image based on the target phase error.
[0010] In some embodiments, the step of inputting the to-be-calibrated phase image and the random perturbation phase image into a target convolutional neural network model to determine the target phase error of the to-be-calibrated phase includes:
[0011] Perform grayscale processing on the to-be-calibrated phase image and the random perturbation phase image respectively to obtain a to-be-calibrated phase grayscale image and a random perturbation phase grayscale image;
[0012] Based on a preset sampling strategy, perform sampling and segmentation on the to-be-calibrated phase grayscale image and the random perturbation phase grayscale image respectively, and then input them into the target convolutional neural network model;
[0013] Through the non-linear mapping relationship between the far-field image and the phase in the target convolutional neural network model, and based on the sampled and segmented to-be-calibrated phase grayscale image and random perturbation phase grayscale image, predict the target phase error of the to-be-calibrated phase image; the non-linear mapping relationship between the far-field image and the phase is obtained by training based on the pair of far-field training images.
[0014] In some embodiments, the optical phased array system further includes a beam splitting component, a lens component and an optical phased array chip; the method further includes:
[0015] When a laser light source generates an output beam after passing through the optical phased array, the output beam is divided into a first beam and a second beam by the beam splitting component;
[0016] Focus the first beam through the lens component to obtain the far-field training image corresponding to the first beam;
[0017] Determine the image electrical signal corresponding to the second beam through the optical phased array chip; based on a preset perturbation strategy, randomly perturb the phase of the image electrical signal corresponding to the second beam to obtain an image perturbed electrical signal corresponding to the second beam;
[0018] Determine the second beam far-field training image and the second beam far-field random perturbation phase image corresponding to the image electrical signal and the image perturbed electrical signal respectively.
[0019] In some of these embodiments, the preset convolutional neural network model includes a residual module and a sampling module; the method further includes:
[0020] Input the far-field training image pair with the phase mapping range restricted and the corresponding random phase distribution of the far-field training image pair into the preset convolutional neural network;
[0021] Extract features from the far-field training image pair through the residual module to obtain training image features;
[0022] Downsample the training image features through the sampling module to output target training image features; the size of the target training image features is set to half of the training image features;
[0023] Determine the predicted phase error according to the target training image features;
[0024] Based on a preset normalized evaluation function, determine the target convolutional neural network model according to the predicted phase error and the true compensation phase corresponding to the far-field training image pair.
[0025] In some of these embodiments, the determining the target convolutional neural network model based on a preset normalized evaluation function according to the predicted phase error and the true compensation phase corresponding to the far-field training image pair includes:
[0026] Calculate the phase difference between the predicted phase error and the true compensation phase;
[0027] When it is determined that the phase difference is a multiple of the first phase, input the far-field training image pair and the corresponding random phase distribution of the far-field training image pair into the preset convolutional neural network model for training until the phase difference is a multiple of the second phase;
[0028] When it is determined that the phase difference is a multiple of the second phase, determine the training stop time in the preset convolutional neural network model, and after the current time reaches the training stop time, determine the target convolutional neural network model; the second phase is greater than the first phase.
[0029] In some of these embodiments, the determining the target convolutional neural network model includes:
[0030] Train the preset convolutional neural network model according to a preset mean square error loss function to obtain the target convolutional neural network model.
[0031] In some of these embodiments, the calibrating the initial emission electric field of the phase image to be calibrated based on the phase error further includes:
[0032] Determine the initial emission electric field that generates the phase image to be calibrated;
[0033] Calibrate the initial emission electric field according to the phase error to obtain the target emission electric field;
[0034] Based on a preset far-field intensity calculation strategy, determine the far-field intensity of the target emission electric field after phase calibration according to the phase error;
[0035] When it is determined that the far-field intensity of the target emission electric field meets a preset intensity threshold, generate a target phase image according to the target emission electric field; the target phase image includes an image generated after calibrating the image to be calibrated.
[0036] In a second aspect, an optical phased array phase calibration system is provided in this embodiment. The system includes: an input module, an error determination module, and a calibration module;
[0037] The input module is configured to input a phase image to be calibrated into the optical phased array system, and obtain a random perturbation phase image corresponding to the phase image to be calibrated; the phase image to be calibrated is generated according to the initial emission electric field in the optical phased array; the random perturbation phase image is obtained by performing phase perturbation within a preset range on the phase image to be calibrated;
[0038] The error determination module is configured to input the phase image to be calibrated and the random perturbation phase image into a target convolutional neural network model, and determine the target phase error of the phase image to be calibrated; the target convolutional neural network model is obtained by training a preset convolutional neural network model based on a pair of far-field training images; the pair of far-field training images includes a far-field training image generated according to the optical phased array and a far-field random perturbation phase image;
[0039] The calibration module is configured to calibrate the initial emission electric field of the phase image to be calibrated based on the target phase error.
[0040] In a third aspect, an electronic device is provided in this embodiment, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the optical phased array phase calibration method described in the first aspect above is implemented.
[0041] In a fourth aspect, a storage medium is provided in this embodiment, on which a computer program is stored. When the program is executed by a processor, the optical phased array phase calibration method described in the first aspect above is implemented.
[0042] Compared with the related art, in an optical phased array phase calibration method, device, electronic device, and storage medium provided in this embodiment, a convolutional neural network is trained through a far-field diffraction image and an image of random phase perturbation to obtain a non-linear mapping relationship between the far-field diffraction image and the phase, so that the trained target convolutional neural network model can avoid symmetric phases in the optical phased array, thereby avoiding the problem of multiple phase solutions. At the same time, by iteratively training the target convolutional neural network model with loss, the accuracy of determining the phase error to be compensated can be effectively improved, and thus the accuracy of the phase error calibration of the optical phased array can be improved.
[0043] Details of one or more embodiments of the present application are set forth in the following drawings and description to make other features, objects, and advantages of the present application more concise and understandable. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments and descriptions thereof of the present application are used to explain the present application and do not constitute an improper limitation of the present application. In the drawings:
[0045] Figure 1 is a hardware structure block diagram of a terminal of the optical phased array phase calibration method provided by an embodiment of the present application;
[0046] Figure 2 is a flowchart of the optical phased array phase calibration method provided by an embodiment of the present application;
[0047] Figure 3 is a flowchart of determining a far-field training image pair in an embodiment of the present application;
[0048] Figure 4 is a schematic structural diagram of an optical phased array system provided by an embodiment of the present application;
[0049] Figure 5 is a structural block diagram of the optical phased array phase calibration system of this embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0050] To understand the purpose, technical solution, and advantages of the present application more clearly, the present application will be described and illustrated below with reference to the drawings and embodiments.
[0051] Unless otherwise defined, the technical terms or scientific terms involved in this application shall have the general meanings understood by those with ordinary skills in the technical field to which this application belongs. In this application, words such as "a", "an", "one kind", "the", "these", etc. do not indicate a limitation in quantity, and they can be singular or plural. The terms "include", "comprise", "have" and any variants thereof involved in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product or device that includes a series of steps or modules (units) is not limited to the listed steps or modules (units), but may include unlisted steps or modules (units), or may include other steps or modules (units) inherent in these processes, methods, products or devices. The terms "connected", "linked", "coupled", etc. involved in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The term "plurality" involved in this application means two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" may represent: A exists alone, A and B exist simultaneously, and B exists alone. Usually, the character " / " indicates that the objects associated before and after are in an "or" relationship. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific order for the objects.
[0052] The method embodiments provided in this embodiment may be executed on a terminal, a computer or a similar computing device. For example, when running on a terminal, Figure 1 is a block diagram of the hardware structure of the terminal of the optical phased array phase calibration method provided by the embodiments of this application. As Figure 1 shown, the terminal may include one or more ( Figure 1 only one is shown in the figure) processors 102 and a memory 104 for storing data. Among them, the processor 102 may include, but is not limited to, a processing device such as a microprocessor MCU or a field programmable gate array FPGA. The above terminal may also include a transmission device 106 for communication functions and an input / output device 108. Those of ordinary skill in the art can understand that Figure 1 the structure shown is only schematic and does not limit the structure of the above terminal. For example, the terminal may also include more or fewer components than those shown in Figure 1 the figure, or have a different configuration from that shown in Figure 1 the figure.
[0053] The memory 104 can be used to store computer programs, such as software programs and modules of application software, such as the computer program corresponding to the optical phased array phase calibration method in this embodiment. The processor 102 executes various functional applications and data processing by running the computer programs stored in the memory 104, that is, implements the above method. The memory 104 may include a high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memories. In some instances, the memory 104 may further include a memory remotely disposed relative to the processor 102, and these remote memories can be connected to the terminal through a network. Examples of the above networks include but are not limited to the Internet, intranet, local area network, mobile communication network, and combinations thereof.
[0054] The transmission device 106 is used to receive or send data via a network. The above network includes a wireless network provided by the communication provider of the terminal. In one instance, the transmission device 106 includes a network adapter (Network Interface Controller, abbreviated as NIC), which can be connected to other network devices through a base station and thus communicate with the Internet. In one instance, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.
[0055] The core of the optical phased array (OPA) technology lies in synthesizing a directional light beam or directionally receiving an incident light beam by controlling the phases of a series of closely arranged optical transmitting or receiving elements. However, for a large-element optical phased array, processing deviations will inevitably occur during the processing, resulting in a phase error in the output optical field of the transmitting antenna unit. Currently, in order to calibrate the phase error and achieve high-precision deflection of the OPA output light beam, many important algorithms have been proposed to calibrate the output optical field, such as the genetic algorithm (GA), hill climbing algorithm (HC), and stochastic parallel gradient descent algorithm (SPGD), etc.
[0056] Among them, the hill-climbing algorithm is a local search algorithm used to find local optimal solutions. It starts from an initial solution and then tries to find a better solution through a series of improvement steps. If a new solution is better than the current one, the algorithm will move to this new solution and continue the search. This process is repeated until no better solution can be found. However, it is difficult to calibrate large-scale OPAs using the hill-climbing algorithm because it scans channel electrical signals one by one, resulting in too low time efficiency. The genetic algorithm can be used to calibrate large-scale operational amplifiers, but its convergence time is too long (feedback cycle, long phase compensation time), and the efficiency of phase calibration needs to be improved urgently. The stochastic parallel gradient descent algorithm is applicable to the combination of continuous waves and high-repetition-rate pulses, but the number of steps required for convergence (feedback cycle) is about ten times the number of combined beams. If there is one sampling point for each pulse and one step or less for each sampling point, then for the combination of pulses with a repetition rate of thousands of hertz and hundreds of beams, this process becomes too slow. The stochastic parallel gradient descent algorithm is hindered by obtaining information from only one detector, unable to indicate which beam or which direction to adjust, and it also introduces noise through the dithering and search processes. Moreover, the traditional stochastic parallel gradient descent algorithm is prone to getting stuck in local optimal solutions, resulting in a slower convergence speed.
[0057] Meanwhile, for an optical phased array, its array arrangement is regular. Taking the sub-beam at the center position as the origin, an initial phase value symmetric about 180° in space can be obtained. For the phase values symmetric about the center and the same phase difference will be generated, resulting in the same far-field image. In the image recognition of CNN, there is a non-unique phase value in the obtained far-field image. As the input of the model, the far-field image cannot accurately reflect the phase information of the OPA, and the CNN model will also be unable to predict a phase value to match the same optimal solution. At the same time, it cannot guarantee that the training model learns the mapping relationship from the far-field image to the optical phase. For different symmetric arrays, different array phases will generate the same far-field image. In practice, to distinguish different phase values, such as phase solution value 1 and phase solution value 2, usually an additional beam-splitting device (wavefront modulator and beam splitter) is adopted, but this will increase the system integration difficulty. Especially for large-array optical phased arrays, it is not the optimal solution. Due to the central symmetry of the optical phase space of the optical phased array, there may be a situation of non-unique solutions during phase calibration, and the current phase calibration algorithms are insufficient to meet the requirements of OPA beam phase calibration.
[0058] Therefore, in this embodiment, an optical phased array phase calibration method is provided, which is applied to an optical phased array system. The optical phased array system includes an optical phased array and a target convolutional neural network model; by adopting the structure of a non-uniform array to distinguish two unique solutions of the array phase, the problem of phase ambiguity is solved. On the one hand, during the process of optimizing the array structure, the array structure is trained and optimized to achieve the problem of low grating lobes of the optical phased array; on the other hand, according to the far-field training images including the original image and the randomly perturbed phase image, the non-uniform array structure between the trained far-field image and the phase can avoid the existence of conjugate symmetric phases and avoid the problem of phase ambiguity. Figure 2 is a flowchart of the optical phased array phase calibration method provided by the embodiment of the present application, as Figure 2 shown, the process includes the following steps:
[0059] Step S210, input the phase image to be calibrated into the optical phased array system, and obtain a randomly perturbed phase image corresponding to the phase image to be calibrated; the phase image to be calibrated is generated according to the initial emission electric field in the optical phased array; the randomly perturbed phase image is obtained by performing phase perturbation within a preset range on the phase image to be calibrated.
[0060] During the process of far-field imaging realized by the laser light source via the optical phased array, if phenomena such as beam offset or misalignment, or abnormal beam shape occur, at this time, it indicates that there is a phase error in the phase image to be calibrated. Therefore, it is necessary to determine the specific array phase solution, that is, the phase error, and calibrate the array phase of the initial emission electric field according to the phase error, and then calibrate the generated phase image to be calibrated.
[0061] When it is necessary to calibrate the phase image to be calibrated, the phase image to be calibrated needs to be input into the optical phased array system; the optical phased array system includes a target convolutional neural network model trained according to the far-field training image and the far-field randomly perturbed phase image, which contains the non-linear mapping relationship between the far-field image and the phase.
[0062] After inputting the phase image to be calibrated into the optical phased array system, obtain the phase of the initial emission electric field that generates the phase image to be calibrated, and according to the phase perturbation within a preset range, obtain the randomly perturbed phase image generated by the initial emission electric field after phase perturbation. Here, the phase perturbation within the preset range can be determined according to the historical phase error, or can be determined according to the array phases of the actual optical phased array and the laser light source, and no specific limitation is made here.
[0063] The optical phased array system here may include an optical phased array, a phase control module for controlling the phase of the optical phased array, and a lens assembly for subsequent imaging, etc., and no specific limitation is made here.
[0064] In step S220, the phase image to be calibrated and the random perturbation phase image are input into the target convolutional neural network model to determine the target phase error of the phase to be calibrated. The target convolutional neural network model is obtained by training a preset convolutional neural network model based on far-field training image pairs. The far-field training image pairs include far-field training images generated according to the optical phased array and far-field random perturbation phase images.
[0065] In this step, after obtaining the phase image to be calibrated and the random perturbation phase image with random phase perturbation, the phase image to be calibrated and the random perturbation phase image are input into the target convolutional neural network model. The target phase error is obtained by predicting the phase error of the phase image to be calibrated according to the target convolutional neural network model.
[0066] Among them, convolutional neural networks (CNNs) are a class of feedforward neural networks with convolutional calculations and deep structures. The general architecture of CNNs usually includes an input layer, a convolutional layer, an activation function, a pooling layer, a fully connected layer, an output layer, etc. In some architectures, a normalization layer is also included to accelerate the training process and stabilize the gradient. The preset convolutional neural network model is determined based on the above convolutional neural network.
[0067] In step S230, the initial emission electric field of the phase image to be calibrated is calibrated based on the target phase error.
[0068] In this step, after the phase image to be calibrated is input into the optical phased array system, the target phase error of the phase image to be calibrated is determined according to the trained target convolutional neural network model. The array phase of the initial emission electric field forming the phase image to be calibrated is calibrated according to the target phase error to obtain the calibrated target emission electric field.
[0069] Through the above steps, by using far-field training image pairs including far-field training images generated according to the optical phased array and far-field random perturbation phase images, that is, far-field training image pairs including far-field diffraction images and random phase perturbations, the preset convolutional neural network model is trained to obtain a target convolutional neural network model including the non-linear mapping relationship between the far-field diffraction image and the corresponding phase, so that the trained target convolutional neural network model can avoid the symmetric phase in the optical phased array, thereby avoiding the problem of multiple phase solutions. At the same time, through the loss iterative training of the target convolutional neural network model, the accuracy of determining the phase error to be compensated can be effectively improved, and thus the accuracy of the phase error calibration of the optical phased array can be improved.
[0070] In some of these embodiments, calibrating the initial emission electric field of the phase image to be calibrated based on the target phase error further includes: determining the initial emission electric field that generates the phase image to be calibrated; calibrating the initial emission electric field according to the target phase error to obtain the target emission electric field; based on a preset far-field intensity calculation strategy, determining the far-field intensity of the target emission electric field after phase calibration according to the target phase error; when it is determined that the far-field intensity of the target emission electric field meets the preset intensity threshold, generating a target phase image according to the target emission electric field; the target phase image includes the image generated after calibrating the image to be calibrated.
[0071] Among them, the preset far-field intensity calculation strategy includes being determined based on the element factor of the target emission electric field, the phase of the optical path in the element, and the target phase error that needs to be compensated, and no specific limitation is made here. When the calculated far-field intensity of the target emission electric field meets the preset intensity threshold, it indicates that there is no phase error in the target emission electric field at this time. Through the optical phased array, a far-field spot with more concentrated far-field energy can be generated.
[0072] In some of these embodiments, the acquisition method of the far-field training image pair for training the target convolutional neural network model can be determined in the optical phased array system based on a preset perturbation strategy; the optical phased array system includes an optical phased array and a target convolutional neural network model, and also includes a beam splitting component, a lens component, and an optical phased array chip; the chip of the optical phased array here is regulated by a phase control module to achieve far-field imaging. Exemplarily, the phase control module is not shown in the figure and can be an FPGA (Field-Programmable Gate Array), a DSP (digital signal processor), an MCU (Microcontroller Unit), etc., and no specific limitation is made here; the optical phased array chip here is specifically set as a silicon-based optical phased array chip.
[0073] When the laser light source passes through the optical phased array regulated by the phase control module, an output beam is generated, and the far-field training image pair is determined according to the output beam.
[0074] Specifically, Figure 3 is the flowchart for determining the far-field training image pair in the embodiments of the present application. Referring to Figure 3 , the method for determining the far-field training image pair includes the following steps S310 to step S340.
[0075] Step S310, when the laser light source generates an output beam after passing through the optical phased array, the output beam is divided into a first beam and a second beam by the beam splitting component.
[0076] Step S320: Focus the first light beam through the lens assembly to obtain the far-field training image corresponding to the first light beam.
[0077] Among them, according to the lens assembly to focus the first light beam to obtain the far-field training image, determine the performance index of the target convolutional neural network model according to the far-field training image, and then continuously correct and train the preset convolutional neural network model to obtain the target convolutional neural network that meets the performance index.
[0078] Step S330: Determine the image electrical signal corresponding to the second light beam through the optical phased array chip; based on the preset perturbation strategy, randomly perturb the phase of the image electrical signal corresponding to the second light beam to obtain the image perturbation electrical signal corresponding to the second light beam.
[0079] Among them, the preset perturbation strategy can be to add Gaussian noise to the image electrical signal corresponding to the second light beam, and the Gaussian noise here is determined based on the phase error existing in the actual imaging process.
[0080] Step S340: Determine the second light beam far-field training image and the second light beam far-field random perturbation phase image corresponding to the image electrical signal and the image perturbation electrical signal respectively.
[0081] By dividing the output light beam generated by the laser light source passing through the optical phased array into two light beams, and respectively performing the processing of focusing and perturbation training on the two light beams, the finally trained target convolutional neural network is made more robust, thereby improving the phase error of the light beam array elements in the optical phased array and obtaining a stable output emission electric field.
[0082] Furthermore, the method for determining the phase image to be calibrated and the random perturbation phase image is the same as the method for determining the far-field training image pair, and will not be described again here.
[0083] In some of the embodiments, input the phase image to be calibrated and the random perturbation phase image into the target convolutional neural network model to determine the target phase error of the phase to be calibrated, including: respectively performing gray processing on the phase image to be calibrated and the random perturbation phase image to obtain the gray phase image to be calibrated and the gray random perturbation phase image; based on the preset sampling strategy, respectively perform sampling and segmentation on the gray phase image to be calibrated and the gray random perturbation phase image and then input them into the target convolutional neural network model; through the non-linear mapping relationship between the far-field image and the phase in the target convolutional neural network model, and based on the sampled and segmented gray phase image to be calibrated and the gray random perturbation phase image, predict the target phase error of the phase image to be calibrated; the non-linear mapping relationship between the far-field image and the phase is trained according to the far-field training image pair.
[0084] By performing grayscale processing on the image, the data volume and computational complexity can be significantly reduced. By sampling and segmenting the image, the image resolution can be reduced, the amount of data to be processed can be decreased, thereby accelerating the processing speed and reducing the demand for computational resources. By performing grayscale processing and sampling segmentation methods on the phase image to be calibrated and the random perturbation phase image, more features in the image can be extracted, which is beneficial to further improving the accuracy of determining the non-linear mapping relationship between the far-field image and the phase, and further improving the accuracy and efficiency of predicting the target phase error subsequently.
[0085] During the process of training a preset convolutional neural network model based on the far-field training image pair, grayscale processing and sampling segmentation are performed on the far-field training image pair, and then the non-linear mapping relationship between the far-field image and the phase is obtained through training.
[0086] In some of the embodiments, the preset convolutional neural network model includes a residual module and a sampling module; the above optical phased array phase calibration method further includes: inputting the far-field training image pair with restricted phase mapping range and the corresponding random phase distribution of the far-field training image pair into the preset convolutional neural network; extracting features of the far-field training image pair through the residual module to obtain training image features; performing downsampling on the training image features through the sampling module to output target training image features; the size of the target training image features is set to be half of the training image features; determining the predicted phase error according to the target training image features; and determining the target convolutional neural network model based on the preset normalized evaluation function according to the predicted phase error and the true compensation phase corresponding to the far-field training image pair.
[0087] Among them, determining the target convolutional neural network model based on the preset normalized evaluation function according to the predicted phase error and the true compensation phase corresponding to the far-field training image pair includes: calculating the phase difference between the predicted phase error and the true compensation phase; when it is determined that the phase difference is a multiple of the first phase, inputting the far-field training image pair and the corresponding random phase distribution of the far-field training image pair into the preset convolutional neural network model for training until the phase difference is a multiple of the second phase; when it is determined that the phase difference is a multiple of the second phase, determining the training termination time in the preset convolutional neural network model, and after the current time reaches the training termination time, determining the target convolutional neural network model; the second phase is greater than the first phase.
[0088] Among them, the first phase can be an odd multiple phase of π, and the second phase is an even multiple phase of π. Exemplarily, the first phase is π, and the second phase is 2π.
[0089] In some of these embodiments, determining the target convolutional neural network model includes: training a preset convolutional neural network model according to a preset mean square error loss function to obtain the target convolutional neural network model.
[0090] The following describes and illustrates this embodiment through specific examples.
[0091] In this specific embodiment, a fast convergence algorithm for calibrating the phase error of beam deflection of a large array optical phased array is proposed. The optical phased array is controlled by a novel machine learning-based iterative method to calibrate the beam phase with an experimentally calibrated numerical simulation model. The array of the optical phased array here can be 32, 64, 128, etc. Here, an example of a 64-array is set. Based on the recognition of the far-field interference pattern of the beam of the optical phased array, a neural network is trained to detect the phase error. There may be non-unique solutions in the optical phase space, and the neural network needs to be trained within a limited phase perturbation, which can reduce the number of samples required for training. When targeting the phase with a small phase perturbation, the trained neural network can make the phase value of the emitted electric field in the optical phased array within the range of 2π, and its iterative scheme can be randomly detected and can converge quickly within the training range. To solve the problem of non-unique solutions in the prior art, double-frame far-field imaging images are used as the input data set to improve the generalization ability of the CNN and make the initial phase compensation result better.
[0092] Convolutional Neural Network (CNN) is a relatively powerful image processing tool, and the expressive ability brought by its diverse architectures has the potential to fit complex mapping relationships. The convolutional kernel in its network structure is a two-dimensional structure, and the weights between network parameters are shared, which can enhance the feature learning ability and computational efficiency of the neural network for image data. Applying CNN to the optical phased array phase calibration system to achieve fast phase locking between array beams, a large number of far-field images with phase labels can be used to train CNN, which can accurately establish the non-linear mapping relationship between the far-field diffraction image and the system phase, and can directly predict the current phase error inside the system.
[0093] The above CNN technology can be used to achieve fast phase retrieval and calibration in the OPA system. Compared with the traditional SPGD algorithm commonly used at present, this technology has great advantages: (1) Compared with the SPGD algorithm, more steps are required for a lower performance evaluation function to recover the aberration. Compared with CNN, it has a better retrieval effect for different initial results, and can obtain a higher far-field output in one step and is more robust to different input ranges; (2) CNN is a wavefront sensor and can directly predict the OPA aberration without additional iterative optimization; (3) When facing an OPA with a large array structure, only the parameters of the output layer need to be fine-tuned to adapt to the number of sub-beams, and its convergence time can remain unchanged.
[0094] The above steps help to precisely control the behavior of the optical phased array to achieve the desired far-field performance, such as applications like beam focusing or shaping in a specific direction. Through continuous iteration and optimization, this algorithm gradually reduces the difference between the actual output and the target, thereby achieving the optimization goal.
[0095] Furthermore, the residual neural network method is adopted. Its skip connection structure, namely the shortcut structure, can effectively avoid gradient dispersion / explosion and suppress network degradation. Due to the existence of shortcut connections, even when the network is very deep, the gradient can be effectively backpropagated to the previous layers through these direct paths. This avoids the common problem of vanishing gradients in traditional deep networks because the gradient no longer needs to be passed layer by layer in a decreasing manner. Network degradation refers to the phenomenon that as the network depth increases, the performance may instead become worse. ResNet overcomes this challenge by allowing the network to learn the residual mapping instead of the complete underlying mapping, making the learning process easier. A network constructed using residual blocks can achieve better accuracy than a network without residual blocks under the same conditions.
[0096] The target convolutional neural network model in this embodiment is based on the basic residual neural network module: among them, the conventional residual module extracts features from the input and does not change the size (height and width) of the feature map during the calculation process; another residual module adopted in this embodiment performs downsampling during the feature extraction process, and the size of the output feature is half of the input feature. The shortcut connection of the residual network can propagate the features extracted from the shallow layer to the deep layer. While enhancing the expression ability of the neural network through feature reuse, it avoids the risk of overfitting.
[0097] In this embodiment, features are extracted from the input layer by adopting an optimized neural network architecture and downsampled. The shortcut connection of the residual network can extend the features extracted from the shallow layer to the deep layer, thereby enhancing the expression ability of the neural network through feature reuse.
[0098] Among them, the scaling factor depends on the target requirements and hardware resource constraints. To retain more detailed information, a smaller scaling factor may be adopted. However, as the scaling factor increases, although the model size can be effectively reduced, it may also lead to the loss of important information, thereby affecting the final accuracy. Exemplarily, a bilinear sampling with a scaling factor of 0.5 is used in the downsampling module.
[0099] Furthermore, batch normalization is used in the convolutional neural network model to make the training more stable. Here, Swish is selected as the activation function because it has a smooth non-linearity and shows better performance in terms of stable training and improved accuracy. The expression of the Swish activation function is as follows:
[0100] ;
[0101] where x represents the input value, and σ(x) is the Sigmoid function; to transform the non-linear problem into a linear problem, x represents the input value of the activation function, which is the result of the weighted input sum received by the current neuron plus the bias term.
[0102] Furthermore, the convolutional neural network model is converged by setting a neural network evaluation function. The OPA phase calibration performance is evaluated by compensating for the phase error, and then the performance of the CNN under different parameters is compared to determine the training termination time. In view of the characteristics of the array beam aberration, in this embodiment, the normalized phase difference cosine is used to evaluate the network performance, and the neural network evaluation function has the formula:
[0103] ;
[0104] where N represents the number of beam arrays of the optical phased array, represents the phase to be compensated, represents the phase predicted during the neural network optimization process.
[0105] When the smaller indicates that the prediction of the neural network for the far-field phase of the beam is closer to the test actual value. The value range of is [0, 1]. The smaller this value, the closer the predicted value of the network is to the true value. When the phase and is an integer multiple of 2π, and will be 0, and the predicted value is equal to the true value. When the difference between is an odd multiple of π,
[0106] In one specific embodiment, the method for training a neural network includes: using 20,000 sets of OPA far-field imaging pictures with a random phase distribution as the input source of the neural network, where each set of pictures includes an initial image and a random phase perturbation image. 15,000 sets of far-field images and their corresponding random phases (phase mapping range is [0, 2π]) are used for training, and the remaining 5,000 sets are used for testing. Further, the mean squared error (MSE) is used as the loss function to train our network, and the average value of the evaluation function is monitored as it changes with the training steps. Exemplarily, a test is conducted every 10 training steps. Through the above neural network evaluation function describes the piston phase residual. During the actual experimental process, after 260 epochs of training, the phase error can be reduced to 0.005. The lower the value, the closer the network output value is to the actual value, the more accurate the network output, and the more accurate the image generated after phase error compensation. The neural network here is the convolutional neural network model in the foregoing embodiment.
[0107] The optical phased array system includes a field-programmable gate array (FPGA), an analog-to-digital converter (ADC), a digital-to-analog converter (DAC), an electro-optic phase modulator, and an acousto-optic modulator, etc.; it also involves techniques such as pseudo-random bit sequence (PRBS), frequency hopping (FH), and heterodyne frequency. Among them, PRBS is a commonly used test signal pattern, which can provide a standardized method to evaluate the performance of the system, such as the bit error rate, during the system test and calibration process. Frequency hopping is a radio communication technology that avoids interference and eavesdropping by quickly switching the operating frequency to optimize the quality and security of the communication link. The heterodyne technique involves mixing the received signal with a reference signal generated by a local oscillator to produce a lower-frequency signal for subsequent processing. In a heterodyne receiver, this process allows for effective amplification and filtering of high-frequency signals.
[0108] Figure 4 is a schematic structural diagram of the optical phased array system provided by the embodiments of the present application. Refer to Figure 4, To verify the ability of the CNN to calibrate the phase, an optical phased array system with 36 elements was constructed to generate a dataset. The encapsulated 64-element (8×8) OPA chip was adjusted by the FPGA phase control module to achieve far-field imaging. After the laser light source was input into the FPGA phase control module, the light beam output by the OPA was divided into two light beams by a beam splitter. One part of the light beam, i.e., the first light beam, passed through a focusing lens with a focal length of 10 cm to the focal plane of the photodetector, which was used to generate the CNN performance index, detect the far-field image for continuous training, and finally used to correct random dynamic aberrations. The other part of the light beam, i.e., the second light beam, was used to obtain the dataset of images with newly assigned random phase perturbations. In this way, 200 groups of datasets were generated. The second light beam generated the corresponding image electrical signal through a charge-coupled device, and then, via a programmable array, such as an FPGA, based on a preset perturbation strategy, randomly perturbed the phase of the image electrical signal corresponding to the second light beam to obtain the image perturbation electrical signal corresponding to the second light beam. Gray processing and sampling segmentation were performed on the second light beam far-field training image and the second light beam far-field randomly perturbed phase image corresponding to the image electrical signal and the image perturbation electrical signal, respectively. As shown in the figure, the sizes of the images obtained by sampling segmentation were 32×32×2, 16×16×64, 4×4×256, and 2×2×512 in sequence; then, the second light beam far-field training image and the second light beam far-field randomly perturbed phase image after sampling segmentation were respectively input into a preset convolutional neural network model to be trained or a trained target convolutional neural network model for model training or model application. Among them, P1-P6 respectively represent the outputs of multiple layers in the convolutional neural network model.
[0109] Different from the far-field image simulated by the first light beam, in the dataset corresponding to the second light beam, including the second light beam far-field training image and the second light beam far-field randomly perturbed phase image, there may be the influence of phase noise, which brings the distribution difference between the experimental data and the training data and affects the performance of the model. Therefore, by sampling the size of the input image during training and adding Gaussian noise to the initial image to simulate noise, the training model is made more robust. Finally, the beam error in the OPA system can be corrected extremely quickly, and a stable and good light field is finally obtained.
[0110] Compared with using the approximate linear relationship between the mean square of the aberration gradient and the second moment of the far-field intensity distribution to process the far-field distribution gray image, in this embodiment, the CNN is used to subsample the OPA far-field gray image to extract more features. For the deterministic mapping problem, the phase is deduced from the far-field coherent diffraction model to quickly find the phase error to be corrected.
[0111] For the recognition of the far-field diffraction pattern, a convolutional neural network is used to discriminate the mapping function between the far-field diffraction image and the beam phase error. The mapping function can be realized by training the neural network with an interference imaging pattern with known phase perturbations. After identifying the mapping function, the neural network can identify the diffraction pattern of the original beam and find the corresponding phase error for any given pattern within a very short deployment.
[0112] Exemplarily, a neural network architecture for mapping the phase of an 8×8 array beam to a 16×16 far-field diffraction image is constructed. With 256 neurons as an input layer, including 64 predicted beam phases connected to many input terminals, an output signal is generated when the input terminal exceeds the set threshold. Mathematically, a neuron is the sum of all input signals multiplied by a weight matrix that describes the mapping function. The weight matrix determines the sensitivity of the neuron to each input. The training of the network is to adjust the sensitivity of the neuron to each of its inputs. The neuron sensitivity refers to the degree of influence (variation degree) of the random phase of the array on the far-field image, that is, the variation degree of the neural network evaluation function. When training the network, the weights of each neuron in the network are adjusted to change the output of the neural network to match the true output (i.e., the known phase perturbation).
[0113] Based on the trained target convolutional neural network model, after determining the target phase error of the phase image to be calibrated and performing calibration, an array beam diffraction model of the optical phased array beam control system can be established to judge whether the phase error is eliminated after compensation through the target phase error and to determine the far-field intensity after compensation.
[0114] Among them, when establishing an array beam diffraction model of the optical phased array beam control system, the transmitting electric field of the optical phased array can be expressed as:
[0115] ;
[0116] Among them, , λ represents the wavelength of the light wave, m represents the number of array elements in the optical phased array, is the optical field amplitude, is the far-field observation distance, the phase of the array element at m is , the radiation pattern of a single array element is represented by the array element factor represents, represents the incident angle, represents the direction angle.
[0117] Generally, when studying an array, it is usually assumed that the array element spacing d and the amplitude A = 1 of a uniform array. Then, the transmitting electric field of the above optical phased array can be simplified as:
[0118] ;
[0119] Wherein, m represents the number of array elements in the optical phased array, is the far-field observation distance, = md, the phase of the array element at m is , and the radiation pattern of a single array element is represented by the array element factor . represents the incident angle, represents the direction angle.
[0120] When there is a phase error, it is necessary to compensate the phase of the optical field. Then, the compensated far-field intensity can be expressed as:
[0121] ;
[0122] Where: is the phase noise of the m-th optical path, is the phase to be compensated. represents the phase of the m-th optical path.
[0123] In an actual system, when there is no phase noise, .
[0124] When the above equation does not hold, it indicates that there is actual phase noise. Then, it is necessary to compensate for the phase noise of the optical field of the optical phased array to obtain a far-field spot with more concentrated far-field energy.
[0125] In this embodiment, an optical phased array phase calibration system is further provided. This system is used to implement the above-mentioned embodiment and the preferred implementation manner, and the descriptions that have been given will not be repeated here. The following terms such as "module", "unit", "sub-unit", etc. can be a combination of software and / or hardware that can achieve a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation in hardware, or a combination of software and hardware, is also possible and contemplated.
[0126] Figure 5 is the structural block diagram of the optical phased array phase calibration system of this embodiment. As Figure 5 shown, this system includes: an input module 10, an error determination module 20, and a calibration module 30.
[0127] The input module 10 is configured to input the phase image to be calibrated into the optical phased array system to obtain a random perturbation phase image corresponding to the phase image to be calibrated; the phase image to be calibrated is generated according to the initial emission electric field in the optical phased array; the random perturbation phase image is obtained by performing phase perturbation within a preset range on the phase image to be calibrated;
[0128] An error determination module 20 is configured to input a phase image to be calibrated and a random perturbation phase image into a target convolutional neural network model to determine the target phase error of the phase to be calibrated. The target convolutional neural network model is obtained by training a preset convolutional neural network model based on a pair of far-field training images. The pair of far-field training images includes a far-field training image generated according to an optical phased array and a far-field random perturbation phase image.
[0129] A calibration module 30 is configured to calibrate the initial emission electric field of the phase image to be calibrated based on the target phase error.
[0130] It should be noted that each of the above modules can be a functional module or a program module, and can be implemented either by software or by hardware. For the modules implemented by hardware, each of the above modules can be located in the same processor; or each of the above modules can also be located in different processors in any combined form.
[0131] In this embodiment, an electronic device is further provided, including a memory and a processor. A computer program is stored in the memory, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.
[0132] Optionally, the above electronic device may further include a transmission device and an input / output device. Among them, the transmission device is connected to the above processor, and the input / output device is connected to the above processor.
[0133] Optionally, in this embodiment, the above processor may be configured to execute the following steps by a computer program:
[0134] S1. Input the phase image to be calibrated into an optical phased array system to obtain a random perturbation phase image corresponding to the phase image to be calibrated. The phase image to be calibrated is generated according to the initial emission electric field in the optical phased array. The random perturbation phase image is obtained by performing phase perturbation within a preset range on the phase image to be calibrated.
[0135] S2. Input the phase image to be calibrated and the random perturbation phase image into the target convolutional neural network model to determine the target phase error of the phase to be calibrated. The target convolutional neural network model is obtained by training a preset convolutional neural network model based on a pair of far-field training images. The pair of far-field training images includes a far-field training image generated according to an optical phased array and a far-field random perturbation phase image.
[0136] S3. Calibrate the initial emission electric field of the phase image to be calibrated based on the target phase error.
[0137] It should be noted that for the specific examples in this embodiment, reference can be made to the examples described in the above embodiments and optional implementation manners, and they will not be elaborated herein.
[0138] In addition, in combination with the optical phased array phase calibration method provided in the above embodiments, a storage medium can also be provided in this embodiment to implement it. A computer program is stored on the storage medium; when the computer program is executed by a processor, any one of the optical phased array phase calibration methods in the above embodiments is implemented.
[0139] It should be understood that the specific embodiments described herein are only used to explain this application, rather than to limit it. According to the embodiments provided in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the protection scope of this application.
[0140] Obviously, the accompanying drawings are only some examples or embodiments of this application. For those of ordinary skill in the art, this application can also be applied to other similar situations based on these drawings without creative efforts. Additionally, it can be understood that although the work done during the development process here may be complex and time-consuming, for those of ordinary skill in the art, certain design, manufacturing, or production changes based on the technical content disclosed in this application are only conventional technical means and should not be regarded as insufficient disclosure of this application.
[0141] The term "embodiment" in this application means that the specific features, structures, or characteristics described in combination with the embodiment can be included in at least one embodiment of this application. The phrase appears in various positions in the specification does not necessarily mean the same embodiment, nor does it mean being independent or alternative to other embodiments and mutually exclusive. Those of ordinary skill in the art can clearly or implicitly understand that the embodiments described in this application can be combined with other embodiments without conflict.
[0142] The above-described embodiments only represent several implementation manners of this application, and their descriptions are relatively specific and detailed, but should not be construed as a limitation on the scope of patent protection. It should be noted that for those of ordinary skill in the art, without departing from the concept of this application, several deformations and improvements can still be made, and these all belong to the protection scope of this application. Therefore, the protection scope of this application should be subject to the appended claims.
Claims
1. An optical phased array phase calibration method, characterized in that, Applied to an optical phased array system, the optical phased array system includes an optical phased array and a target convolutional neural network model; the method includes: Input the phase image to be calibrated into the optical phased array system to obtain a random perturbation phase image corresponding to the phase image to be calibrated; the phase image to be calibrated is generated according to the initial emission electric field in the optical phased array; the random perturbation phase image is obtained by performing phase perturbation within a preset range on the phase image to be calibrated; Input the phase image to be calibrated and the random perturbation phase image into the target convolutional neural network model to determine the target phase error of the phase to be calibrated; the target convolutional neural network model is trained based on far-field training image pairs for a preset convolutional neural network model; the far-field training image pairs include far-field training images generated according to the optical phased array and far-field random perturbation phase images; Calibrate the initial emission electric field of the phase image to be calibrated based on the target phase error.
2. The optical phased array phase calibration method according to claim 1, wherein The step of inputting the phase image to be calibrated and the random perturbation phase image into the target convolutional neural network model to determine the target phase error of the phase to be calibrated includes: Perform grayscale processing on the phase image to be calibrated and the random perturbation phase image respectively to obtain a grayscale image of the phase to be calibrated and a grayscale image of the random perturbation phase; Based on a preset sampling strategy, perform sampling and segmentation on the grayscale image of the phase to be calibrated and the grayscale image of the random perturbation phase respectively, and then input them into the target convolutional neural network model; Through the non-linear mapping relationship between the far-field image and the phase in the target convolutional neural network model, and based on the sampled and segmented grayscale image of the phase to be calibrated and the grayscale image of the random perturbation phase, predict the target phase error of the phase image to be calibrated; the non-linear mapping relationship between the far-field image and the phase is trained according to the far-field training image pairs.
3. The optical phased array phase calibration method according to claim 1, wherein The optical phased array system further includes a beam splitting component, a lens component, and an optical phased array chip; the method further includes: When the laser light source generates an output beam after passing through the optical phased array, divide the output beam into a first beam and a second beam via the beam splitting component; Focus the first beam through the lens component to obtain the far-field training image corresponding to the first beam; Through the optical phased array chip, determine the image electrical signal corresponding to the second beam; based on a preset perturbation strategy, randomly perturb the phase of the image electrical signal corresponding to the second beam to obtain an image perturbation electrical signal corresponding to the second beam; Respectively determine the second beam far-field training image and the second beam far-field random perturbation phase image corresponding to the image electrical signal and the image perturbation electrical signal.
4. The optical phased array phase calibration method according to claim 1, wherein The preset convolutional neural network model includes a residual module and a sampling module; the method further includes: Input the far-field training image pairs with restricted phase mapping range and the random phase distribution corresponding to the far-field training image pairs into a preset convolutional neural network; Extract features from the far-field training image pairs through the residual module to obtain training image features; Downsample the training image features through the sampling module to output target training image features; the size of the target training image features is set to be half of the training image features; Determine the predicted phase error according to the target training image features; Based on a preset normalization evaluation function, determine the target convolutional neural network model according to the predicted phase error and the true compensation phase corresponding to the far-field training image pair; 5. The optical phased array phase calibration method according to claim 4, wherein The determining the target convolutional neural network model based on a preset normalization evaluation function according to the predicted phase error and the true compensation phase corresponding to the far-field training image pair includes: Calculate the phase difference between the predicted phase error and the true compensation phase; When it is determined that the phase difference is a multiple of the first phase, input the far-field training image pair and the random phase distribution corresponding to the far-field training image pair into a preset convolutional neural network model for training until the phase difference is a multiple of the second phase; When it is determined that the phase difference is a multiple of the second phase, determine the training stop time in the preset convolutional neural network model, and after the current time reaches the training stop time, determine the target convolutional neural network model; the second phase is greater than the first phase.
6. The optical phased array phase calibration method according to claim 5, wherein The determining the target convolutional neural network model includes: Train the preset convolutional neural network model according to a preset mean square error loss function to obtain the target convolutional neural network model.
7. The optical phased array phase calibration method according to claim 1, wherein The calibrating the initial emission electric field of the phase image to be calibrated based on the phase error further includes: Determine the initial emission electric field that generates the phase image to be calibrated; Calibrate the initial emission electric field according to the phase error to obtain the target emission electric field; Based on a preset far-field intensity calculation strategy, determine the far-field intensity of the target emission electric field after phase calibration according to the phase error; When it is determined that the far-field intensity of the target emission electric field meets a preset intensity threshold, generate a target phase image according to the target emission electric field; the target phase image includes the image generated after calibrating the image to be calibrated.
8. An optical phased array phase calibration system, characterized in that, The system includes: an input module, an error determination module, and a calibration module; The input module is configured to input the phase image to be calibrated into the optical phased array system to obtain a random perturbation phase image corresponding to the phase image to be calibrated; the phase image to be calibrated is generated according to the initial emission electric field in the optical phased array; the random perturbation phase image is obtained by performing phase perturbation within a preset range on the phase image to be calibrated; The error determination module is configured to input the phase image to be calibrated and the random perturbation phase image into the target convolutional neural network model to determine the target phase error of the phase image to be calibrated; the target convolutional neural network model is obtained by training a preset convolutional neural network model based on a far-field training image pair; the far-field training image pair includes a far-field training image and a far-field random perturbation phase image generated according to the optical phased array; The calibration module is used to calibrate the initial transmitted electric field of the phase image to be calibrated based on the target phase error.
9. An electronic device, comprising a memory and a processor, characterized in that, A computer program is stored in the memory, and the processor is configured to run the computer program to execute the optical phased array phase calibration method according to any one of claims 1 to 7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, the steps of the optical phased array phase calibration method according to any one of claims 1 to 7 are implemented.
Citation Information
Patent Citations
Optical phased array calibration system and method based on neural network
CN115169213A
Optical phased array initial phase calibration system and method based on convolutional neural network
CN117055273A
Photonic isin computation engine with optical phased array
CN118541658A
Initial phase calibration method of optical phased array
CN118625573A
Phased array phase deviation calibration method and device, equipment and storage medium
CN118777996A
Cited By
Data transmission method and system based on optical phased array and program product
CN122419623A
Large-scale optical phased array phase mixing calibration method
CN122429934A
A large-scale optical phased array phase mixing calibration method
CN122429934B
