An optical phased array based data transmission method, system and program product

Abstract

The application provides an optical phased array-based data transmission method, system and program product. The method comprises: obtaining a target beam parameter; obtaining a target phase prediction model based on optical phase physical property constraints and migration learning training; processing the target beam parameter through the target phase prediction model to obtain a target phase distribution; converting the target phase distribution into a driving voltage based on an independent calibration mapping relationship of each channel; dynamically compensating the driving voltage based on real-time state monitoring results; and driving a phase shifter based on the compensated driving voltage to control beam deflection. The application solves the problems of low phase control precision and poor environmental adaptability, and realizes high-precision, fast switching and long-term stable beam control.

Description

Technical Field

[0001] This application relates to the field of optical communication technology, specifically to a data transmission method, system, and program product based on an optical phased array. Background Technology

[0002] In large data centers and high-performance computing clusters, optical phased arrays (OPA) are increasingly being used in optical interconnect systems to achieve flexible data transmission due to their advantages such as having no moving mechanical parts and fast switching speed.

[0003] However, in practical engineering applications, OPA chips manufactured using silicon-based optoelectronic processes inevitably suffer from manufacturing errors such as waveguide width and thickness deviations, leading to random phase errors in each channel. Simultaneously, under prolonged high-load operation, the phase modulation characteristics of thermo-optical phase shifters change due to thermal crosstalk and temperature drift. Existing phase correction methods typically rely on purely data-driven neural networks or iterative search algorithms. The former often ignores the physical characteristics of optical phase (such as periodicity and translation invariance), resulting in far-field confusion during training and a high dependence on real data; the latter has a slow convergence speed, making it difficult to meet the requirements of microsecond-level link establishment. Furthermore, existing solutions lack fine-grained calibration for individual hardware differences and real-time compensation mechanisms for dynamic environmental changes, leading to decreased beam pointing accuracy, deteriorated sidelobe suppression ratio, and consequently, increased data transmission error rate and insufficient link stability.

[0004] Therefore, there is an urgent need for a data transmission method, system, and software product based on optical phased arrays. Summary of the Invention

[0005] In view of the above-mentioned problems in related technologies, this application provides a data transmission method, system and program product based on optical phased array.

[0006] The objective of this application is achieved through the following technical solution:

[0007] In a first aspect, this application provides a data transmission method based on an optical phased array, comprising:

[0008] Obtain the target beam parameters and the target phase prediction model, wherein the target phase prediction model is a model obtained by physical characteristic constraints of optical phase and transfer learning training;

[0009] The target beam parameters are processed using the target phase prediction model to obtain the target phase distribution;

[0010] Based on the independent calibration mapping relationship of each channel, the target phase distribution is converted into the driving voltage of each channel;

[0011] Based on the real-time status monitoring results, each of the driving voltages is dynamically compensated to obtain the corresponding compensated driving voltage.

[0012] The phase shifter of the optical phased array is driven based on the compensated driving voltage to control the beam deflection.

[0013] Preferably, the real-time status monitoring result includes a link health score; the link health score is determined based on a set of link performance indicator information, which includes first performance indicator information and second performance indicator information.

[0014] The link health score includes a preliminary score determined based on the second performance indicator information, and a subsequent score determined based on the first performance indicator information when the preliminary score exceeds the reference threshold range.

[0015] The dynamic compensation strategy is determined based on the link health score level. The dynamic compensation strategy includes: maintaining regular electrical compensation when the link health score is at a normal level; enhancing electrical compensation and shortening the status monitoring cycle when the link health score is at a slightly deteriorated level; and triggering thermal compensation and initiating a recalibration process when the link health score is at an abnormal level.

[0016] Preferably, the step of dynamically compensating each of the driving voltages based on real-time status monitoring results includes:

[0017] When monitoring data for the target channel is missing within a predetermined time period, the historical performance index information of the target channel and the real-time performance index information of adjacent channels are input into a lightweight time series prediction model to obtain the inferred phase compensation value.

[0018] The driving voltage of the target channel is corrected based on the inferred phase compensation value so that the target channel maintains a degraded operating state.

[0019] Preferably, the method further includes:

[0020] During data transmission idle time slots, the optical phased array is driven to emit test beams based on known test beam parameters;

[0021] The measured feedback data of the test beam is collected, and the measured feedback data is compared with the pre-stored reference feedback data to obtain the model drift amount;

[0022] When the model drift exceeds a preset update threshold, the parameters of the first few layers of the target phase prediction model are frozen, the parameters of the last few layers and the output layer are fine-tuned using the measured feedback data, and / or a rapid recalibration of the independent calibration mapping relationship is triggered.

[0023] Preferably, the known test beam parameters correspond to a preset fixed deflection angle or a preset scanning trajectory; the step of comparing the measured feedback data with pre-stored reference feedback data to obtain the model drift includes:

[0024] Calculate the angular deviation between the measured far-field main lobe position and the reference main lobe position, as well as the difference between the measured sidelobe suppression ratio and the reference sidelobe suppression ratio, and quantitatively evaluate the output drift of the target phase prediction model.

[0025] By comparing the deviation between the actual driving voltage and the theoretical voltage of each channel, the drift of the independent calibration mapping relationship is evaluated.

[0026] The triggering of rapid recalibration of the independent calibration mapping relationship includes: re-performing a sine function fitting scan on channels whose drift exceeds a preset calibration update threshold, and updating their phase shift voltage values ​​and independent calibration mapping relationships.

[0027] Preferably, the independent calibration mapping relationship is a mapping relationship established based on the following method:

[0028] A varying driving signal is applied to the phase shifter of each channel in the optical phased array, and the periodic change in far-field diffraction efficiency is monitored.

[0029] The periodic changes are fitted with a sine function to independently obtain the phase shift voltage value of each channel; based on the phase shift voltage value, a mapping between channel identifier and driving voltage is established.

[0030] Preferably, the target phase prediction model is a model trained using a loss function that includes direction vector constraints and an activation function that matches phase periodicity; wherein, the direction vector constraints are used to eliminate far-field confusion caused by phase translation.

[0031] Secondly, this application also provides a data transmission device based on an optical phased array, including a processor and a memory, wherein the memory stores a computer program, and the computer program, when executed by the processor, implements the data transmission method as described above.

[0032] Thirdly, this application also provides a data transmission system based on an optical phased array, including a signal transmitter, a signal receiver, and a data transmission device as described above; the signal transmitter and the signal receiver each include an optical phased array, and the data transmission device is used to control the beam deflection of the optical phased array to realize data transmission between the signal transmitter and the signal receiver.

[0033] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the data transmission method as described in any one of the first aspects.

[0034] This application provides a data transmission method, system, and program product based on an optical phased array. By incorporating the physical characteristics of optical phase constraints into the training process of the target phase prediction model, it avoids the far-field confusion and solution space ambiguity problems caused by traditional pure data-driven methods that neglect physical priors such as phase translation invariance and periodic boundaries. Simultaneously, by leveraging a transfer learning strategy of simulation pre-training and real data fine-tuning, it significantly reduces the amount of data and time required for real-world calibration while ensuring model prediction accuracy and convergence stability. Furthermore, by calling the independent calibration mapping relationship of each channel to convert the target phase distribution into a driving voltage, it compensates for deviations caused by silicon-based optoelectronic processes. Individual differences in the response characteristics of each phase shifter eliminate systematic wavefront distortion introduced by manufacturing tolerances, effectively suppressing sidelobe level rise and improving beam pointing accuracy. Furthermore, dynamic compensation of the driving voltage based on real-time status monitoring results enables continuous sensing and offsetting of phase modulation characteristic changes caused by environmental factors such as thermal crosstalk and temperature drift during data transmission link maintenance, achieving rapid phase stabilization in speed response and reducing optical power loss and bit error rate caused by pointing deviation. The synergistic effect of these technologies enables optical phased arrays to simultaneously achieve high-precision, rapid switching, and long-term stable beam control under complex operating conditions, significantly improving the data transmission quality and engineering robustness of optical interconnect systems. Attached Figure Description

[0035] The present application will be further described below with reference to the accompanying drawings and embodiments.

[0036] Figure 1 This is a flowchart illustrating a data transmission method provided in an embodiment of this application.

[0037] Figure 2 This is a schematic diagram of a process for dynamically compensating each of the driving voltages provided in an embodiment of this application.

[0038] Figure 3 This is a partial flowchart illustrating a data transmission method provided in an embodiment of this application. Detailed Implementation

[0039] The present application will now be further described in conjunction with the accompanying drawings and specific embodiments. It should be noted that, without conflict, the various embodiments or technical features described below can be arbitrarily combined to form new embodiments. The implementation process of the present application will be described below with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present application from the content disclosed in this specification. The present application can also be implemented or applied through other different specific implementation procedures, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present application. It should be understood that the preferred embodiments are only for illustrating the present application and not for limiting the scope of protection of the present application.

[0040] Example 1

[0041] like Figure 1 As shown, this embodiment provides a data transmission method based on an optical phased array. This method is applied to an optical interconnect unit containing an optical phased array to achieve rapid and precise pointing control of the light beam during data transmission. Specifically, the method includes the following steps:

[0042] Step S101: Obtain the target beam parameters.

[0043] Specifically, target beam parameters are digital descriptions of the desired beam emission state. They can be directly represented by target deflection angles (e.g., horizontal deflection angle θ and vertical deflection angle φ), or by the intensity distribution matrix of the target far-field beam spot or a specific beamforming vector. In data transmission scenarios, these parameters are typically generated and transmitted in real time by the upper-layer routing protocol based on the spatial location of the target receiver, or obtained by querying a locally stored routing lookup table.

[0044] Step S102: Obtain the target phase prediction model, which is a model obtained by physical characteristic constraints of optical phase and transfer learning training.

[0045] In this embodiment, acquiring the target phase prediction model refers to loading and calling a pre-trained and solidified model instance from local memory or external storage medium during the method execution phase. This target phase prediction model is not built or trained on-site during data transmission, but is directly used as a pre-defined component with specific input-output mapping capabilities. The phrase "obtained based on the physical characteristics constraints of optical phase and training through transfer learning" defines the inherent properties of the target phase prediction model, indicating that it has internalized prior physical knowledge such as the periodicity and translation invariance of optical phase before leaving the factory or deployment, and has undergone transfer adaptation between simulation data and real data. During runtime, the processor only needs to load the weight file or inference engine of the target phase prediction model to put it in standby mode, thereby avoiding the huge computational overhead and latency caused by online training and ensuring the real-time establishment of the data transmission link. In specific applications, S101 and S102 can be executed simultaneously or sequentially.

[0046] Step S103: The target beam parameters are processed by the target phase prediction model to obtain the target phase distribution.

[0047] Specifically, as part of the forward inference process, the processor uses the target beam parameters obtained in step S101 as input tensors and feeds them into the loaded target phase prediction model. Based on its internally fixed nonlinear mapping relationship, the model directly outputs the target phase values ​​required for each antenna channel of the optical phased array, thus forming the target phase distribution. For example, for a linear array containing 128 channels, the model output is a 128-dimensional phase vector, with each element corresponding to the ideal phase modulation amount of a channel. Compared to traditional iterative search algorithms (such as genetic algorithms or hill-climbing methods) that require hundreds of iterations to converge, the model inference method used in this embodiment can generate a high-precision phase distribution in one go, significantly improving the beam switching speed and meeting the low-latency requirements of data center optical interconnects.

[0048] Step S104: Based on the independent calibration mapping relationship of each channel, the target phase distribution is converted into the driving voltage of each channel.

[0049] Due to variations in manufacturing processes, the phase shifter response characteristics of each channel in an optical phased array are not identical, making precise phase control impossible with a uniform theoretical voltage. Therefore, this embodiment invokes pre-stored independent calibration mapping relationships for each channel during runtime. These mapping relationships can be either a look-up table established separately for each channel or analytical function coefficients characterizing the phase-voltage response of that channel. During conversion, for each phase value in the target phase distribution, the corresponding channel's dedicated drive voltage is queried or calculated. This one-channel-one-map mechanism eliminates phase errors introduced by individual device differences, ensuring that the applied electrical signal is accurately converted into the desired optical phase modulation.

[0050] Step S105: Based on the real-time status monitoring results, dynamically compensate each of the driving voltages to obtain the corresponding compensated driving voltage.

[0051] In practical applications, optical phased arrays are affected by dynamic factors such as changes in ambient temperature and thermal crosstalk during operation, causing phase shift characteristics to drift. Optical phased array (OPA) chips typically integrate, for example, an optical phased array, a thermo-optical phase shifter (or an electro-optical phase shifter), a temperature sensor (for monitoring the chip's own temperature), a backlight monitoring photodiode (for monitoring output optical power), a digital-to-analog converter (DAC), and a driving circuit. This embodiment uses on-chip sensors (such as a temperature sensor or a backlight monitoring photodiode) to collect real-time status data and corrects the driving voltage generated in step S104 according to a preset compensation strategy. For example, when a rise in chip temperature is detected, the voltage compensation amount is automatically calculated based on a pre-stored temperature-phase drift coefficient and added to the original driving voltage; or, when abnormal fluctuations in output optical power are detected, a closed-loop feedback algorithm is triggered to fine-tune the voltage value. This dynamic compensation process is a closed-loop adjustment that is continuously or periodically executed during the data transmission link maintenance, effectively suppressing the impact of environmental disturbances on beam pointing accuracy and ensuring the stability of long-term transmission.

[0052] Step S106: Drive the phase shifter of the optical phased array based on the compensated driving voltage to control the beam deflection.

[0053] Specifically, the compensated driving voltage is applied to the electrodes of the corresponding thermo-optical or electro-optic phase shifter of the optical phased array via a digital-to-analog converter (DAC) and a driving amplifier circuit. The phase shifter changes the waveguide refractive index according to the voltage, thereby modulating the phase of the transmitted light wave.

[0054] When the phase modulation of all channels reaches the value set by the target phase distribution, the emitted light from each channel coherently superimposes in the far field, forming a high-gain main lobe beam pointing in the direction of the target, thereby realizing the precise transmission or reception of optical signals to the designated receiving end.

[0055] Therefore, this embodiment, through the complete execution flow of S101 to S106 described above, incorporates the physical characteristics constraints of the optical phase into the training process of the target phase prediction model. This avoids the far-field confusion and solution space ambiguity problems caused by neglecting physical priors such as phase translation invariance and periodic boundaries in traditional pure data-driven methods. Simultaneously, by leveraging a transfer learning strategy of simulation pre-training and real data fine-tuning, the amount of data and time required for actual calibration is significantly reduced while ensuring model prediction accuracy and convergence stability. Furthermore, by calling the independent calibration mapping relationship of each channel to convert the target phase distribution into a driving voltage, it compensates for the phase variations caused by deviations in silicon-based optoelectronic processes. Individual differences in shifter response characteristics eliminate systematic wavefront distortion introduced by manufacturing tolerances, effectively suppressing sidelobe level rise and improving beam pointing accuracy. Furthermore, dynamic compensation of the driving voltage based on real-time status monitoring results enables continuous sensing and offsetting of phase modulation characteristic changes caused by environmental factors such as thermal crosstalk and temperature drift during data transmission link maintenance. The speed response achieves rapid phase stabilization, reducing optical power loss and bit error rate caused by pointing deviation. The synergistic effect of these technologies enables optical phased arrays to simultaneously achieve high-precision, rapid switching, and long-term stable beam control under complex operating conditions, significantly improving the data transmission quality and engineering robustness of optical interconnect systems.

[0056] In one specific embodiment, the internal structure and properties of the target phase prediction model are further defined.

[0057] Specifically, the target phase prediction model is a model trained using a loss function that includes direction vector constraints and an activation function that matches phase periodicity. The direction vector constraints are used to eliminate far-field confusion caused by phase translation. In the physical mechanism of optical phased arrays, when the phase of all channels increases or decreases by the same constant simultaneously, the far-field intensity distribution remains unchanged; this phenomenon is called global phase translation invariance.

[0058] Specifically, the direction vector constraint can be implemented using the following loss function term:

[0059]

[0060] Where N is the total number of channels in the optical phased array. The predicted phase of the nth channel. Let be the label phase of the nth channel. This loss function can utilize the natural periodicity of the cosine function, so that the loss value is zero when the predicted phase differs from the label phase by 2kπ (k is an integer), thereby eliminating the ambiguity in the solution space caused by the global translation component; at the same time, it only depends on the phase difference rather than the absolute phase value, ensuring the consistency between the optimization objective and the far-field physical distribution.

[0061] If only the traditional mean squared error (MSE) is used as the loss function, the model will encounter serious ambiguity in the solution space during training. That is, for the same far-field target, there are infinitely many phase solutions that satisfy the conditions, which leads to oscillations in the gradient descent direction and makes it difficult to converge to the optimal solution with physical realizability.

[0062] To address this issue, the directional vector constraint introduced in this application is mathematically a regulation of the relative relationship of the phase distribution. For example, this constraint can be manifested by converting the predicted phase and the label phase into unit complex vectors and then calculating their dot product similarity, or by calculating the cosine of the phase difference between the two as part of the loss term. In this way, the optimization objective of the loss function is consistent with the physical characteristics of the phase distribution, thereby eliminating the interference from the global translation component and ensuring that the phase distribution output by the model is physically deterministic and unique.

[0063] Furthermore, to adapt to the cyclic characteristics of the phase, the target phase prediction model employs an activation function that matches the phase periodicity. Specifically, since the optical phase has a periodic boundary of 0 to 2π, conventional monotonic activation functions such as ReLU or Sigmoid will produce numerical abrupt changes or discontinuities at the boundary, disrupting the topological continuity of the phase. This embodiment preferably uses the Sin activation function or other periodic activation units based on trigonometric functions, enabling the output layer of the neural network to naturally possess the ability to perform cyclic mapping from 0 to 2π. This allows the target phase prediction model to directly output phase values ​​that conform to the physical definition without additional post-processing modulo operations, simplifying the computational chain during inference and avoiding gradient truncation problems caused by modulo operations, significantly improving the prediction accuracy and stability of the target phase prediction model at the phase boundary.

[0064] As one implementation method, the target phase prediction model is a model obtained by pre-training an initial neural network based on simulated far-field data with random phase errors and then fine-tuning it using real far-field data.

[0065] It is important to emphasize that during the data transmission method operation phase of this application, the target phase prediction model is already a finished product after completing the aforementioned pre-training and fine-tuning processes. Its network weights are fixed or updated only under specific triggering conditions, rather than being trained in real time for each beam control. Specifically, the simulated far-field data is generated by mixing data with different phase error variance ratios. Specifically, to ensure the model's broad adaptability to OPA chips of varying quality, a single error level is not used when constructing the simulation dataset; instead, multiple error distributions are mixed according to a preset ratio. For example, data samples with phase error standard deviations σ=π, 0.75π, and 0.5π can be mixed in a 3:2:1 ratio. This mixing strategy allows the model to learn error patterns ranging from severe distortion to slight perturbations during the pre-training phase, establishing robust far-field-phase inverse mapping prior knowledge and preventing the model from overfitting to a specific ideal or extreme error state.

[0066] During fine-tuning, the parameters of the first few fully connected layers of the initial neural network are frozen, and only the parameters of the last few layers and the output layer are fine-tuned. This parameter state is a specific projection of the transfer learning strategy onto the model entity. Specifically, the first few layers of the neural network are usually responsible for extracting general optical features of the far-field pattern (such as main lobe position, side lobe envelope, etc.). These features are highly common across different OPA chips, so they are frozen during the fine-tuning stage to maintain their generalization ability. The last few layers and the output layer are mainly responsible for adapting to the unique high-frequency phase noise caused by the manufacturing tolerance of a specific chip, so they are kept adjustable to absorb personalized information from real data. This parameter configuration of freezing before adjusting avoids catastrophic forgetting caused by small sample real data and achieves accurate compensation for individual differences. The real far-field data consists of multiple sets of far-field patterns collected within a preset field of view. For example, dozens to hundreds of actual emitted light spot patterns can be collected using an infrared camera or photodetector array within a field of view of -12° to 12° as a fine-tuning reference. Compared to the massive amounts of data required for training from scratch, this model building method based on physical constraints and transfer learning reduces the data requirements for real-machine calibration, enabling each optical interconnect unit to complete high-precision model adaptation within minutes, whether at the factory or in the deployment site.

[0067] In one specific embodiment, the basis for constructing the independent calibration mapping relationship and its fault-tolerance mechanism under non-ideal hardware conditions are described in detail. The independent calibration mapping relationship is established based on the following method:

[0068] A varying driving signal is applied to the phase shifter of each channel in the optical phased array, and the periodic change in far-field diffraction efficiency is monitored. The periodic change is fitted by a sine function to independently obtain the phase shift voltage value of each channel. Based on the phase shift voltage value, a mapping between channel identifier and driving voltage is established.

[0069] In silicon-based optoelectronic processes, phase shifters typically operate based on thermo-optical effects or carrier dispersion effects. Taking the most widely used thermo-optical phase shifter as an example, its working principle involves changing the waveguide temperature using a metal heater, thereby altering the refractive index using the material's thermo-optical coefficient, ultimately modulating the phase of the transmitted light. Since the optical waveguide itself constitutes an interference structure (such as the arms of a Mach-Zehnder interferometer or a micro-ring resonator), there is a natural trigonometric function relationship between the output light intensity or far-field diffraction efficiency and the applied phase delay, exhibiting a periodic oscillation in the form of a sine or cosine. Therefore, fitting the monitored diffraction efficiency data with a sine function provides a precise mathematical description of the phase shifter's physical response characteristics, rather than a simple empirical approximation. In contrast, linear or polynomial fitting not only fails to cover a wide range of phase modulations exceeding 2π but also introduces significant nonlinear errors near extreme points, leading to a decrease in beam control accuracy.

[0070] In the specific implementation process, scanning and fitting operations are performed independently for each channel. For example, for the nth channel, the processor controls the digital-to-analog converter to output N voltage sampling points uniformly distributed within the range from 0V to the maximum safe voltage, while simultaneously recording the far-field diffraction efficiency value fed back by the photodetector. Subsequently, the acquired data sequence is fitted into a sine curve using the least squares method or the Levenberg-Marquardt algorithm. The angular frequency ω directly reflects the voltage-to-phase conversion efficiency of the phase shifter for that channel, from which the precise voltage value required to achieve the π phase shift (i.e., ...) can be calculated. Due to waveguide width deviations, doping concentration fluctuations, and heater resistance differences during the manufacturing process, different channels... Values ​​often exhibit significant dispersion. Through the aforementioned independent calibration, a unique voltage-phase lookup table or analytical coefficients can be generated for each channel. This allows for the accurate conversion of the target phase distribution into a driving voltage that compensates for individual differences during subsequent real-time data transmission, fundamentally eliminating systematic wavefront distortion caused by process tolerances.

[0071] Furthermore, considering the potential defects or aging failures in actual hardware, this application also introduces an automatic identification and isolation mechanism for abnormal channels. As one implementation, the process of establishing the independent calibration mapping relationship further includes: during the process of fitting the periodic change using a sine function, if the far-field diffraction efficiency of the target channel does not exhibit a preset periodic change characteristic, then the target channel is identified as an abnormal channel; subsequently, when converting the target phase distribution into the driving voltage of each channel, the driving voltage corresponding to the abnormal channel is fixed to a preset constant bias voltage, and the phase contribution of the abnormal channel is removed from the target phase distribution.

[0072] This mechanism ensures the reliable operation of the optical phased array system under non-ideal hardware conditions. The absence of a pre-defined periodic variation characteristic can be defined in practical engineering using various quantitative criteria. For example, a goodness-of-fit threshold can be set; if the coefficient of determination R² of the sinusoidal fit is below 0.8, the channel response is considered abnormal. Alternatively, an amplitude threshold can be set; if the amplitude A of the fitted sinusoidal wave is less than three times the background noise level, the channel may be in an open-circuit or severely damaged state. The monotonicity of the detection curve can also be considered; if no obvious extreme point is detected within the expected period, it can be deemed a failure. These criteria can be used individually or in combination to improve detection accuracy. It should be understood that the specific threshold values ​​mentioned above are merely illustrative and can be adjusted according to the system's signal-to-noise ratio requirements and device specifications in practical applications.

[0073] Once a channel is identified as abnormal, the system immediately initiates a dual isolation strategy. First, at the hardware driver level, the driving voltage of that channel is locked to a preset constant bias voltage. This bias voltage is typically chosen to be a static operating point that puts the phase shifter in a low-loss or thermally stable state. The purpose is to avoid unpredictable thermal crosstalk or electrical noise introduced by floating or random jumps in the driving signal, preventing interference with adjacent normal channels. Second, at the algorithm control level, when calculating the target phase distribution or performing beamforming, the weighting factor of the abnormal channel is set to zero, or the term is directly removed from the array factor summation formula. This means that when generating the beam, the system actively "ignores" the existence of this channel, relying on the coherent superposition of the remaining normal channels to maintain the beam pointing. Although the reduction in effective aperture may slightly widen the main lobe width or slightly increase the sidelobe level, compared to the wavefront confusion and pointing deviation caused by forcibly driving a bad point with an unknown response, this conscious elimination strategy can maximize the determinism of beam quality and the availability of the link. This hardware-software co-design significantly improves the optical phased array data transmission system's ability to adapt to manufacturing defects and long-term aging.

[0074] In other implementations, if the phase shifter uses other physical mechanisms such as electro-optic modulation, its calibration function may also be adjusted to other periodic functions that conform to the physical law. Similarly, the handling of abnormal channels is not limited to fixed bias and elimination, but may also include alternative solutions such as redundant channel replacement or deweighting.

[0075] In one specific embodiment, based on real-time status monitoring results, dynamic compensation is performed on each of the driving voltages to obtain the corresponding compensated driving voltage, including:

[0076] The chip temperature of the optical phased array is monitored in real time. When the change in chip temperature exceeds a set threshold, the driving voltage is dynamically adjusted based on a preset temperature-phase drift relationship, and / or the temperature control component is activated to perform temperature control compensation on the optical phased array.

[0077] Specifically, this dynamic compensation mechanism is a dedicated correction method designed for the unique physical characteristics of thermo-optical phase shifters in optical phased arrays, rather than general ambient temperature control. In silicon-based optoelectronic devices, the refractive index change of thermo-optical phase shifters is strongly correlated with temperature, and there is a significant thermal crosstalk effect between adjacent channels, resulting in a complex nonlinear drift of the phase modulation amount with chip temperature. If only traditional PID temperature control algorithms are used to maintain the chip constant temperature, not only is the response speed slow and the energy consumption high, but it is also difficult to eliminate the phase error caused by transient thermal disturbances outside the bandwidth of the temperature control loop. Therefore, this application introduces a direct electrical compensation path based on the "temperature-phase drift relationship," which directly maps the temperature change amount to the phase correction amount, thereby achieving active cancellation of thermal drift.

[0078] The method of dynamically adjusting the driving voltage based on a preset temperature-phase drift relationship includes: triggering a local phase correction model, outputting a phase compensation value based on the current chip temperature and the temperature-phase drift relationship, and correcting the driving voltage based on the phase compensation value.

[0079] In this approach, the temperature-phase drift relationship is an inherent parameter characterizing the thermo-optical response of the specific optical phased array chip. Specifically, it can be a linear drift coefficient k (usually in rad / °C or V / °C) that has been experimentally determined and stored in advance, or it can be a polynomial fitting curve or a two-dimensional lookup table describing the nonlinear response.

[0080] For example, during hardware and software initialization or factory calibration, the rate of change of the π phase shift voltage of each channel phase shifter can be measured at different temperature points to construct an accurate drift model. During operation, when the temperature sensor integrated on the chip detects a change ΔT in the current temperature T relative to the reference temperature T0, the local phase correction model immediately calculates the required phase compensation value based on the aforementioned drift relationship. This is then converted into a corresponding voltage correction ΔV and superimposed on the original driving voltage. The response time of this electrical compensation method is limited only by the bandwidth of the digital-to-analog converter and the driving circuit, typically reaching the microsecond or even nanosecond level, which is much faster than the thermal control loop. It can effectively suppress high-frequency phase noise caused by instantaneous power consumption fluctuations or environmental airflow disturbances during data transmission, ensuring the real-time stability of the beam pointing.

[0081] Alternatively, when the change in chip temperature exceeds a set threshold, the calibration results are periodically re-called to correct the independent calibration mapping relationship, and the driving voltage is recalculated based on the corrected independent calibration mapping relationship.

[0082] This path constitutes the system's second safeguard mechanism, primarily used to address significant temperature drift exceeding the linear range of electrical compensation or reference offset caused by long-term aging. Specifically, when the monitored chip temperature change ΔT exceeds a preset safety threshold (e.g., ±5°C or ±10°C), or when the accumulated error of electrical compensation exceeds the tolerance, relying solely on voltage correction may introduce new deviations due to model extrapolation failure.

[0083] At this point, a deeper level of compensation will be triggered: on the one hand, temperature control components (such as a thermoelectric cooler (TEC) or a micro-heater) can be activated to actively compensate for the temperature of the optical phased array, pulling the chip temperature back to the optimal operating range; on the other hand, during business downtime or maintenance windows, pre-stored calibration data can be retrieved or a rapid calibration process (such as a simplified sine fitting) can be executed online to update and correct the independent calibration mapping relationship described in the above embodiments. For example, the best-matching set can be selected from multiple pre-stored calibration tables based on the current temperature as the new mapping reference, or the voltage values ​​in the existing lookup table can be batch-corrected based on the temperature drift model. Through this dual-path collaborative strategy of "electrical compensation as the main method and thermal compensation and recalibration as auxiliary methods," microsecond-level rapid phase stabilization under normal operating conditions is ensured, while the system can automatically recover to a high-precision reference state under extreme operating conditions, avoiding the poor reliability of the thermo-optical phased array in long-term operation under complex thermal environments.

[0084] It should be understood that the specific values ​​of the aforementioned thresholds, the mathematical expression of the temperature-phase drift relationship, and the switching logic between electrical and thermal compensation can all be adaptively adjusted according to the actual optical phased array chip's process parameters, heat dissipation conditions, and the accuracy requirements of the application scenario. For example, in scenarios extremely sensitive to latency, the trigger threshold for thermal compensation can be appropriately relaxed to prioritize ensuring the continuity of electrical compensation; while in metrology-grade applications with extremely high absolute accuracy requirements, the recalibration frequency can be increased to ensure the real-time accuracy of the mapping relationship. These variations are all reasonable extensions of the dynamic compensation mechanism protected in this application.

[0085] In one specific embodiment, the method further includes:

[0086] In the phase shifter arrangement direction orthogonal to the optical phased array, the beam emitted by the optical phased array is controlled to scan in two-dimensional space by wavelength tuning of the tunable laser in conjunction with the compensated driving voltage.

[0087] Specifically, a two-dimensional beam control mechanism based on the synergistic effect of electrically controlled phase and wavelength dispersion was constructed. In the physical architecture of an optical phased array, phase shifters are typically arranged linearly along a single direction (defined as the φ direction), and beam deflection in that direction can be achieved by adjusting the phase difference of each channel. However, a one-dimensional phase shifter array alone cannot directly form a wavefront gradient in an orthogonal direction (defined as the θ direction). This embodiment utilizes the dispersion characteristics of the grating antenna or waveguide itself integrated in the optical path to convert the change in the wavelength of the light source into a change in the diffraction angle in the θ direction. According to the grating equation, when the incident light wavelength changes, the exit angle of its far-field main lobe will shift linearly or non-linearly. Therefore, by precisely controlling the output wavelength of the tunable laser, continuous scanning of the beam in the orthogonal direction can be achieved without adding an additional one-dimensional phase shifter array. This hybrid scanning architecture significantly reduces the wiring complexity of the chip and the number of control channels.

[0088] In this process, the wavelength tuning controls the deflection angle of the beam in the orthogonal direction, and the compensated driving voltage controls the deflection angle of the beam in the phase shifter arrangement direction. In this coordinated control process, the control quantities in the two dimensions are not completely independent, but rather inherently physically coupled.

[0089] Specifically, when the wavelength changes to achieve θ-axis deflection, the φ-axis phase distribution, which was originally calibrated for a specific wavelength, will drift because the optical path difference is inversely proportional to the wavelength. This causes the beam's direction in the φ-axis to deviate from the expected target. To solve this problem, the "coordination" described in this embodiment is not only time synchronization but also includes decoupling and compensation in the control logic. In actual operation, the "compensated driving voltage" obtained in step S105 not only corrects for temperature drift but may also include phase pre-compensation for the current operating wavelength.

[0090] For example, the target phase prediction model can use the current wavelength value as an additional input feature, or it can have a pre-stored reference phase offset table for different wavelengths. When the laser switches to a new wavelength λ... new When calculating the φ-direction driving voltage, a voltage equal to (λ) is automatically superimposed. new -λ refThe relevant phase correction term cancels out the φ-direction crosstalk caused by wavelength changes. This decoupling control ensures that the deflection accuracy of the φ and θ directions does not interfere with each other during two-dimensional scanning, achieving orthogonal independent addressing.

[0091] To illustrate the two-dimensional scanning mechanism more clearly, we will use a specific data center optical interconnect scenario as an example below.

[0092] In this scenario, the optical phased array employs a silicon-based grating antenna array, with phase shifters arranged horizontally. The tuning range of the tunable laser is set to 1500nm to 1600nm, corresponding to a scanning range of -12° to +12° in the θ direction; the scanning range in the horizontal φ direction is also -12° to +12°, precisely controlled by 128 channels of thermo-optical phase shifters using the data transmission methods described in Examples 1 to 4. When it is necessary to switch the beam from point A (θ=-10°, φ=5°) to point B (θ=8°, φ=-3°), the processor first calculates the required target wavelength (e.g., 1565nm) based on the target θ angle and instructs the laser to tune to that wavelength; simultaneously, the processor calls the target phase prediction model, inputting the target φ angle (-3°) and the current wavelength (1565nm), and the model outputs a target phase distribution that includes wavelength decoupling compensation; subsequently, after conversion via independent calibration mapping and dynamic temperature compensation, the final driving voltage is obtained and applied to the phase shifters.

[0093] It should be understood that the above wavelength range and angle values ​​are merely illustrative examples. In other embodiments, the correspondence between the tuning range and the scanning field of view can be adjusted according to the grating period, waveguide refractive index and application requirements. As long as it follows the technical concept of wavelength control orthogonal dimension, phase control arrangement dimension and the two are decoupled in a coordinated manner, it should be covered within the protection scope of this application.

[0094] In one specific embodiment, unlike the above embodiments, the real-time status monitoring result includes a link health score; the link health score is determined based on a set of link performance indicator information, which includes first performance indicator information and second performance indicator information.

[0095] Specifically, the second performance indicator information characterizes the end-to-end macroscopic transmission quality, such as the received optical power, bit error rate (BER), and signal-to-noise ratio (SNR) fed back from the receiver; the first performance indicator information characterizes the microscopic physical state of each channel, such as at least one of the following: real-time temperature of each channel, backlight monitoring optical power, driving voltage drift, actual power consumption of the channel, and resistance drift value. Through this hierarchical set of indicators, the operational status of the optical phased array can be grasped from both the communication effect and the physical underlying dimensions, avoiding compensation lag or misjudgment caused by relying solely on a single temperature variable.

[0096] The link health score includes a preliminary score determined based on the second performance indicator information, and a subsequent score determined based on the first performance indicator information when the preliminary score exceeds the reference threshold range.

[0097] Specifically, during data transmission, a preliminary score reflecting end-to-end communication quality is calculated first. For example, the current bit error rate and signal-to-noise ratio are compared with a benchmark value using a weighted mean square error (MSE) method. Only when the preliminary score indicates that the link quality fluctuates or exceeds a preset normal reference threshold range is a deeper acquisition of the first performance indicator information for each channel and a subsequent score calculation triggered. The technical effect of this cascaded triggering mechanism is that it significantly reduces monitoring overhead. During most normal operating periods, it is not necessary to frequently read the underlying sensor data of all channels, thereby freeing up the processor's computing resources for data transmission services. At the same time, when the preliminary score is abnormal, the subsequent score can accurately locate whether it is a deterioration of thermal crosstalk in a specific channel or a global temperature drift, providing a precise basis for subsequent differentiated compensation.

[0098] The dynamic compensation strategy is determined based on the link health score level, and the dynamic compensation strategy includes:

[0099] When the link health score is at a normal level, maintain regular electrical compensation; when the link health score is at a slightly deteriorated level, enhance electrical compensation and shorten the status monitoring cycle; when the link health score is at an abnormal level, trigger thermal compensation and start the recalibration process.

[0100] Specifically, a three-level compensation strategy mapping table can be included to achieve refined adaptive control. When the link health score is at a normal level, conventional electrical compensation is maintained, i.e., basic voltage correction is performed only based on the local phase correction model, and the existing state monitoring cycle is maintained to maintain steady-state operation with minimal power consumption. When the link health score is at a slightly deteriorated level, electrical compensation is enhanced and the state monitoring cycle is shortened, for example, by increasing the compensation gain coefficient to suppress the gradually emerging drift trend, while increasing the monitoring frequency to twice or even more, and optionally activating far-field spot quality auxiliary monitoring to capture more diagnostic information before the link completely fails. When the link health score is at an abnormal level, thermal compensation is triggered and a recalibration process is initiated, for example, by activating the TEC temperature control component for active heat dissipation or heating, while performing rapid recalibration on the independent calibration mapping relationship, and switching to a backup transmission link if necessary. Compared with the traditional binary strategy of alarming upon abnormality, the graded response mechanism adds a buffer zone, enabling flexible intervention in the early stages of failure, significantly improving the robustness and long-term availability of the data transmission link.

[0101] As an example, based on the above scoring mechanism, the link health score level is determined according to the correspondence between the score value and a preset threshold range. Specifically, a first scoring threshold and a second scoring threshold are preset, wherein the first scoring threshold is higher than the second scoring threshold, thereby dividing the link health score into three levels: when the link health score is greater than or equal to the first scoring threshold, it is judged as a normal level, indicating that the overall transmission quality of the link is good and the physical state of each channel is within the allowable deviation range; when the link health score is lower than the first scoring threshold but greater than or equal to the second scoring threshold, it is judged as a slightly degraded level, indicating that the link transmission quality has shown a perceptible downward trend or the physical state parameters of some channels have deviated from the normal range but have not yet reached the level of failure; when the link health score is lower than the second scoring threshold, it is judged as an abnormal level, indicating that the link transmission quality is severely degraded or there is a channel-level physical fault, requiring active intervention measures. For example, in a system with a maximum score of 100, the first scoring threshold can be set to 85 points, and the second scoring threshold can be set to 60 points. That is, a score of 85 points or above is considered normal, a score between 60 and 85 points is considered slightly degraded, and a score below 60 points is considered abnormal. It should be understood that the specific values ​​of the above thresholds can be calibrated and adjusted according to the performance margin of the optical phased array and the system reliability requirements in the actual application scenario, and there is no limitation on this.

[0102] Based on the above-mentioned classification, the dynamic compensation strategy is determined according to the link health score level. Specifically, a three-level compensation strategy mapping table is constructed to achieve refined adaptive control. When the link health score is at the normal level, conventional electrical compensation is maintained, i.e., basic voltage correction is performed only based on the local phase correction model, and the existing state monitoring cycle is maintained to maintain steady-state operation with the lowest power consumption. When the link health score is at the slightly deteriorated level, electrical compensation is enhanced and the state monitoring cycle is shortened, for example, by increasing the compensation gain coefficient to suppress the gradually emerging drift trend, while increasing the monitoring frequency to twice or even more than the original, and optionally starting far-field spot quality auxiliary monitoring to capture more diagnostic information before the link completely fails. When the link health score is at the abnormal level, thermal compensation is triggered and a recalibration process is initiated, for example, by activating the TEC temperature control component for active heat dissipation or heating, while performing rapid recalibration on the independent calibration mapping relationship, and switching to a backup transmission link if necessary.

[0103] See Figure 2 Furthermore, the dynamic compensation of each driving voltage based on real-time status monitoring results includes:

[0104] S201, when the monitoring data of the target channel is missing within a predetermined time period, the historical performance index information of the target channel and the real-time performance index information of the adjacent channels are input into the lightweight time series prediction model to obtain the inferred phase compensation value.

[0105] S202, the driving voltage of the target channel is corrected based on the inferred phase compensation value so that the target channel maintains a degraded operation state.

[0106] Meanwhile, in practical engineering applications, temperature sensors or backlight monitoring photodiodes integrated on the chip may experience temporary data loss due to readout circuit noise or transient interference. If a fixed bias and algorithm-based isolation strategy is directly adopted, the effective aperture of the optical phased array will decrease instantaneously, leading to main lobe broadening, side lobe lifting, and even beam pointing jumps, affecting ongoing communication services. The compensation mechanism proposed in this embodiment utilizes the inherent physical correlation between optical phased array channels: on the one hand, there is thermal crosstalk between adjacent channels, and their temperature and phase changes have a high spatial correlation; on the other hand, the thermal inertia of the chip determines that phase drift is a continuous and gradual temporal process. Therefore, a lightweight timing prediction model (such as an LSTM network or a Transformer encoder) can accurately infer the current true phase drift of a target channel based on a historical performance index sequence of several past cycles, combined with the real-time state of spatially adjacent normal channels.

[0107] The lightweight temporal prediction model employs an encoder-decoder architecture, where the encoder extracts the spatiotemporal features of the input sequence, and the decoder outputs the phase compensation prediction value for the current moment. The encoder can use a single-layer or dual-layer LSTM (Long Short-Term Memory) unit or a small Transformer encoder (e.g., containing 1 to 2 self-attention layers), and the decoder is a fully connected layer. The model's input is a multi-dimensional temporal feature vector, including: a sequence of historical performance indicators for the target channel over the past N sampling periods (e.g., temperature, backlight power, driving voltage drift, etc.), and the current real-time performance indicators of K spatially adjacent normal channels; where N and K are positive integers, for example, N can be 5 to 10, and K can be 2 to 4 (i.e., 1 to 2 adjacent channels to the left and right of the target channel). The model's output is the inferred phase compensation value for the target channel at the current moment, i.e., the predicted phase drift. The overall number of parameters is controlled in the thousands to tens of thousands, and the time taken for a single inference is in the microsecond range. It can be deployed in an embedded processor to achieve real-time online prediction, meeting the microsecond-level beam switching latency requirements of optical phased arrays.

[0108] In terms of model training, as an example, the lightweight time-series prediction model employs a combination of offline pre-training and online incremental updates. During the offline pre-training phase, a training sample set is constructed using synchronous monitoring data from all channels collected during normal operation of the optical phased array system. Performance indicators for each channel over N consecutive sampling periods are used as input features, and the actual phase compensation value for the (N+1)th sampling period of the same channel is used as the label. Simultaneously, the real-time status of adjacent channels is introduced as auxiliary input features. Mean squared error (MSE) or mean absolute error (MAE) is used as the loss function, and the model parameters are optimized through backpropagation. During the online incremental update phase, once the monitoring data for the target channel recovers from a missing state, the system uses the deviation between the inferred phase compensation value output by the model during the previous data loss period and the recovered actual measurement value as a new training sample. The model parameters are then fine-tuned and updated with a preset learning rate, thereby continuously adapting to the drift characteristics evolution caused by device aging and environmental changes.

[0109] While the inferred phase compensation values ​​generated in this way may be slightly less accurate than the actual sensor measurements, they are sufficient to maintain the channel's participation in beamforming, keeping it in a degraded operation rather than a completely offline state, thus avoiding performance spikes caused by momentary failures of a single sensor. Furthermore, once the monitoring data for the target channel returns to normal, it can automatically switch back to a closed-loop compensation mode based on real data, using the deviation between the previously inferred and actual values ​​as new training samples to update the lightweight time-series prediction model online, thereby continuously improving the accuracy of soft compensation. It should be understood that the aforementioned predetermined duration can be flexibly adjusted according to the monitoring sampling period, for example, set to 3 sampling periods; the structure of the lightweight time-series prediction model is not limited to LSTM or Transformer, and any sequence modeling method capable of capturing spatiotemporal correlations can be applied. This combination of hardware and software fault-tolerant design further constructs a dynamic runtime data fault-tolerant defense, enhancing the ability of optical phased array data transmission under complex and non-ideal conditions.

[0110] See Figure 3 In one specific embodiment, the method further includes:

[0111] S107, During the data transmission idle time slot, drive the optical phased array to emit a test beam based on the known test beam parameters;

[0112] S108, Collect the measured feedback data of the test beam, and compare the measured feedback data with the pre-stored reference feedback data to obtain the model drift amount;

[0113] S109, when the model drift exceeds the preset update threshold, freeze the parameters of the first few layers of the target phase prediction model, fine-tune the parameters of the last few layers and the output layer using the measured feedback data, and / or trigger a rapid recalibration of the independent calibration mapping relationship.

[0114] Specifically, an online self-calibration mechanism was constructed to address the gradual performance degradation issue of optical phased arrays and other devices during long-term data transmission without interruption of normal service data transmission. The aforementioned idle data transmission slots refer to time segments within the inter-frame guard interval, routing switching interval, or preset maintenance window period of the optical interconnect link that do not carry valid user data. The controller identifies these slots by monitoring the link status register or receiving upper-layer scheduling instructions and automatically injects the test process within those slots. This non-intrusive triggering mechanism ensures zero interference between the self-calibration process and normal communication services.

[0115] During testing, the known test beam parameters correspond to a preset fixed deflection angle or a preset scanning trajectory. For example, the 0° direction at the center of the field of view and the ±10° directions at the edge of the field of view can be selected as fixed test points, or a low-rate raster scanning trajectory covering the main working area can be set. These test points are usually accurately characterized during the system initialization phase, and their corresponding ideal far-field patterns and driving voltages are stored as pre-stored reference feedback data. By comparing the current measured data with this data, the current performance deviation can be quantitatively assessed.

[0116] Furthermore, the known test beam parameters correspond to a preset fixed deflection angle or a preset scanning trajectory; the step of comparing the measured feedback data with the pre-stored reference feedback data to obtain the model drift includes:

[0117] Calculate the angular deviation between the measured far-field main lobe position and the reference main lobe position, as well as the difference between the measured sidelobe suppression ratio and the reference sidelobe suppression ratio, and quantitatively evaluate the output drift of the target phase prediction model.

[0118] By comparing the deviation between the actual driving voltage and the theoretical voltage of each channel, the drift of the independent calibration mapping relationship is evaluated.

[0119] The triggering of rapid recalibration of the independent calibration mapping relationship includes: re-performing a sine function fitting scan on channels whose drift exceeds a preset calibration update threshold, and updating their phase shift voltage values ​​and independent calibration mapping relationships.

[0120] Among them, the main lobe angle deviation directly reflects whether there is a systematic error in the phase gradient generated by the model, while the deterioration of the sidelobe suppression ratio usually indicates the presence of high-frequency noise in the phase distribution or a decrease in inter-channel consistency. The combination of these two indicators can effectively characterize the degradation of the prediction accuracy of the target phase prediction model under the current operating conditions.

[0121] In addition, the actual driving voltage applied to achieve the target phase was recorded when the test beam was emitted. If there is a significant difference between this actual voltage value and the theoretical voltage value calculated based on the current independent calibration mapping (e.g., exceeding the 3σ statistical threshold), or if there is a monotonically drift compared to the historical voltage value at the last calibration, it indicates that the physical response characteristics (e.g., Vπ value) of the channel phase shifter have changed. This method of separating the optical far-field performance from the electrical drive parameters allows for precise determination of whether the software model needs adjustment or the hardware calibration table needs correction, avoiding the waste of resources caused by blind compensation.

[0122] Based on the above quantitative evaluation results, a differentiated update strategy is implemented. When the output drift of the target phase prediction model is determined to exceed a preset update threshold, online lightweight fine-tuning of the model is triggered. Specifically, the parameters of the first few layers of the target phase prediction model are frozen, and the parameters of the last few layers and the output layer are fine-tuned using the measured feedback data. This strategy can be executed online in real time. Since the first few layers extract general optical features, they usually remain stable in short-term operation, so they are frozen to prevent overfitting or catastrophic forgetting on small sample test data; fine-tuning only the last few layers responsible for adapting to individual differences can quickly absorb the latest drift information with extremely low computational overhead, allowing the model to readjust to the current physical state.

[0123] On the other hand, when an independent calibration mapping relationship is determined to have drifted, a rapid recalibration of the independent calibration mapping relationship is triggered. Specifically, triggering the rapid recalibration of the independent calibration mapping relationship includes: re-performing a sine function fitting scan on channels whose drift exceeds a preset calibration update threshold, updating their phase shift voltage values ​​and independent calibration mapping relationships. It is particularly important to emphasize that this recalibration is local and on-demand. The sine fitting scan process described above is only performed on specific channels diagnosed as having excessive drift, while channels that are still within the normal range maintain their original calibration parameters. Compared to the huge time overhead and service interruption risk caused by the traditional approach of periodically recalibrating all channels globally, this local rapid recalibration strategy reduces calibration time and enables repair during brief idle time slots.

[0124] Through the aforementioned online self-calibration mechanism, a closed-loop system for ensuring accuracy throughout the entire lifecycle is constructed based on the offline training and initial calibration of the above embodiments. It not only automatically detects and compensates for accumulated drift caused by long-term operation, but also achieves an optimal balance between computational resources and calibration accuracy through differentiated update strategies, significantly improving the reliability and availability of optical phased array data transmission under complex long-term operating conditions.

[0125] Example 2

[0126] This embodiment provides a data transmission device and a computer-readable storage medium based on an optical phased array, used to carry and execute the data transmission method described in the foregoing embodiments at the hardware level. Specifically, a data transmission device based on an optical phased array includes a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, it implements the data transmission method as described in any of the foregoing embodiments.

[0127] In this embodiment, the data transmission device is specifically embodied as an integrated circuit, which serves as the control core of the optical interconnect unit and is electrically connected to the OPA transmitter and receiver. The processor can be one or more combinations of a general-purpose processor (such as a CPU), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), or a field-programmable gate array (FPGA). The memory includes volatile memory (such as SRAM, DRAM) and non-volatile memory (such as Flash, EEPROM), both coupled to the processor via a high-speed bus. During actual operation, the computer program is not an abstract logical set but is fixed into a specific sequence of machine instructions. When the processor executes these instructions, it triggers a series of microscopic data flow and control actions: First, the processor loads the weight parameters of the target phase prediction model described in the aforementioned embodiments from non-volatile memory into the on-chip cache or memory; second, after receiving the target beam parameters, the processor calls the arithmetic logic unit (ALU) or tensor computation core to perform matrix operations based on the loaded model weights, and infers the target phase distribution in real time; subsequently, the processor accesses a lookup table stored in a specific address area of ​​memory, which records the independent calibration mapping relationship of each channel described in the aforementioned embodiments, and converts the target phase distribution into an initial driving voltage digital quantity; next, the processor reads the real-time status data fed back by the temperature sensor integrated on the chip, and performs addition or multiplication correction operations on the initial driving voltage digital quantity at the register level according to the temperature-phase drift relationship described in the aforementioned embodiments, generating the final compensated driving voltage data; finally, the processor writes this data into the control register of the digital-to-analog converter (DAC) through the peripheral interface, thereby driving the OPA phase shifter to complete the beam deflection. It should be understood that the above hardware execution process is merely illustrative. In other embodiments, if pure hardware circuits (such as fully custom ASICs) are used to implement part of the algorithm logic, or if a multi-core heterogeneous architecture is used to share inference and control tasks, as long as they can achieve the same data processing function and real-time control effect, they should all be covered within the scope of protection of this application. Through this deeply collaborative hardware and software device architecture, this application transforms complex phase correction algorithms into nanosecond-level electrical signal operations, providing solid physical computing power support for the establishment of microsecond-level links for optical phased arrays in high-frequency switching scenarios such as data centers.

[0128] Furthermore, this application also provides a computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the data transmission method as described in any of the foregoing embodiments. Specifically, the computer-readable storage medium refers to any non-transitory physical medium capable of persistently storing data and being read by a computing device, including but not limited to disks (such as hard disks and floppy disks), optical disks (such as CD-ROMs and DVDs), and semiconductor memories (such as ROMs, EPROMs, flash memory cards, and USB flash drives). The computer program stored on this medium includes all instruction codes, model parameter files, calibration data tables, and configuration files required to implement the data transmission method of this application. When the medium is connected to or installed in the aforementioned data transmission device or other compatible computing device, the stored program instructions are loaded into memory and parsed and executed by the processor, thereby enabling the device to control the beam based on the physical characteristics constraints of optical phase. The significance of this implementation is that it decouples the technical solution of this application from specific hardware binding, allowing it to be independently distributed, sold, or deployed as a software product or firmware update package. For example, during data center operation and maintenance, the phase prediction model or dynamic compensation strategy can be upgraded or optimized without replacing the OPA chip hardware by replacing the storage medium or pushing program updates online, thereby continuously adapting to new business needs or fixing potential defects. It should be understood that although this embodiment lists various specific media forms, this does not constitute a limitation on the scope of protection of this application. Any tangible storage medium capable of carrying the program instructions and being called by the processor falls within the scope of computer-readable storage media as defined in this application.

[0129] Example 3

[0130] This embodiment provides a data transmission system based on an optical phased array, which is an integrated manifestation of the data transmission method described in the foregoing embodiments at the physical device level.

[0131] Specifically, the system includes a signal transmitter, a signal receiver, and a data transmission device as described in the foregoing embodiments. In this embodiment, the signal transmitter and signal receiver are not isolated functional modules, but are specifically configured as multiple optical interconnect units deployed within a data center. Each optical interconnect unit integrates an OPA transmitter, an OPA receiver, and an integrated circuit as the control core.

[0132] The signal transmitting end and the signal receiving end each include an optical phased array, and the data transmission device is used to control the beam deflection of the optical phased array to realize data transmission between the signal transmitting end and the signal receiving end.

[0133] Specifically, the signal transmitter and the signal receiver each independently perform the following steps:

[0134] Obtain the corresponding target beam parameters and target phase prediction models;

[0135] The target beam parameters are processed by their respective target phase prediction models to obtain their respective target phase distributions;

[0136] Based on the independent calibration mapping relationship of each corresponding channel, the target phase distribution is converted into the driving voltage of each channel;

[0137] Based on their respective real-time status monitoring results, each driving voltage is dynamically compensated to obtain the corresponding compensated driving voltage.

[0138] The phase shifters of each optical phased array are driven by the compensated driving voltage to achieve bidirectional beam deflection control between the signal transmitting end and the signal receiving end.

[0139] This architecture means that any optical interconnect unit in the system has both the active beam manipulation capability as a signal transmitter and the high-sensitivity alignment capability as a signal receiver, thus supporting full-duplex and flexible interconnection between any nodes.

[0140] To more clearly illustrate the actual workflow and technical effects of the system in this application, a detailed description is provided below using a data center application scenario. In this scenario, it is assumed that the first optical interconnect unit (as the source unit) needs to transmit a large-scale dataset to the second optical interconnect unit (as the target unit). The entire data transmission link establishment and maintenance process is as follows: First, the integrated circuit of the source unit receives the target address instruction issued by the upper-layer routing protocol and resolves the relative position coordinates of the target unit in three-dimensional space. Subsequently, the source unit calls the target phase prediction model pre-stored in its local memory (i.e., the model trained by physical constraints and transfer learning as described in the aforementioned embodiments), inputs the target deflection angle parameters, and infers the target phase distribution required by each channel of the OPA transmitter. Next, based on the independent calibration mapping relationship described in the aforementioned embodiments, the phase distribution is converted into a precise driving voltage, and dynamic compensation is performed based on the real-time temperature monitoring results described in the aforementioned embodiments, ultimately driving the OPA transmitter to emit a high-gain beam pointing in a specific direction. After the beam's propagation path is changed by an optical modification device (e.g., a high-precision mirror array or holographic optical element), it is precisely projected onto the physical area where the target unit is located.

[0141] Meanwhile, the target unit does not passively wait for the optical signal to arrive, but synchronously executes the same phase control logic as the source unit. Specifically, the target unit's integrated circuit, based on the source unit's orientation information, calls its local target phase prediction model and calibration data to perform reverse beamforming control on the phase shifters of each channel at its OPA receiver, ensuring that the main lobe gain direction of the receiver is precisely aligned with the direction of arrival of the incident optical signal. This bidirectional alignment mechanism, where both the transmitter and receiver cooperate, is a feature that distinguishes this solution from traditional unidirectional illumination or mechanical scanning systems. By employing high-precision phase prediction and dynamic compensation technologies at both ends simultaneously, the link loss caused by insufficient control precision at a single endpoint can be effectively overcome, significantly improving the end-to-end optical power budget margin and signal-to-noise ratio.

[0142] Furthermore, the optical modification device described in this embodiment plays a key role in expanding the interconnect topology in the system. In some implementations, it can also be integrated into the package of the optical interconnect unit or exist as a separate optical backplane. The specific form of the optical modification device is not limited to a planar mirror, but can also be a curved mirror, a metasurface lens, an acousto-optic deflector, or a liquid crystal spatial light modulator, etc. Its core function is to cooperate with the two-dimensional scanning capability of the OPA (such as wavelength-phase co-scanning as described in the embodiment) to relay the light beam to the target node outside the direct field of view of the OPA, thereby achieving full interconnection coverage of hundreds or thousands of storage modules in the data center.

[0143] It should be understood that the specific processes, device selections, and performance indicators for data center interconnection described above are merely illustrative examples intended to help those skilled in the art understand the system-level application value of this application, and are not intended to limit the scope of protection of this application. Any optical system built based on the data transmission device described in this application, regardless of its specific application scenario or optical relay form, as long as it substantially utilizes the phase control and compensation mechanisms described in the foregoing embodiments to achieve beam deflection and data transmission, falls within the scope of the data transmission system based on optical phased arrays claimed in this application.

[0144] Example 4

[0145] This embodiment provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the data transmission method as described in any of the above embodiments.

[0146] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A data transmission method based on an optical phased array, characterized in that, include: Obtain the target beam parameters and the target phase prediction model, wherein the target phase prediction model is a model obtained by physical characteristic constraints of optical phase and transfer learning training; The target beam parameters are processed using the target phase prediction model to obtain the target phase distribution; Based on the independent calibration mapping relationship of each channel, the target phase distribution is converted into the driving voltage of each channel; Based on the real-time status monitoring results, each of the driving voltages is dynamically compensated to obtain the corresponding compensated driving voltage. The phase shifter of the optical phased array is driven based on the compensated driving voltage to control the beam deflection; The real-time status monitoring results include a link health score; The link health score is determined based on a set of link performance indicator information, which includes first performance indicator information and second performance indicator information. The link health score includes a preliminary score determined based on the second performance indicator information, and a subsequent score determined based on the first performance indicator information when the preliminary score exceeds the reference threshold range. The dynamic compensation strategy is determined based on the link health score level. The dynamic compensation strategy includes: maintaining regular electrical compensation when the link health score is at a normal level; enhancing electrical compensation and shortening the status monitoring cycle when the link health score is at a slightly deteriorated level; and triggering thermal compensation and initiating a recalibration process when the link health score is at an abnormal level. The dynamic compensation of each driving voltage based on real-time status monitoring results includes: When monitoring data for the target channel is missing within a predetermined time period, the historical performance index information of the target channel and the real-time performance index information of adjacent channels are input into a lightweight time series prediction model to obtain the inferred phase compensation value. The driving voltage of the target channel is corrected based on the inferred phase compensation value so that the target channel maintains a degraded operating state.

2. The data transmission method according to claim 1, characterized in that, The method further includes: During data transmission idle time slots, the optical phased array is driven to emit test beams based on known test beam parameters; The measured feedback data of the test beam is collected, and the measured feedback data is compared with the pre-stored reference feedback data to obtain the model drift amount; When the model drift exceeds a preset update threshold, the parameters of the first few layers of the target phase prediction model are frozen, the parameters of the last few layers and the output layer are fine-tuned using the measured feedback data, and / or a rapid recalibration of the independent calibration mapping relationship is triggered.

3. The data transmission method according to claim 2, characterized in that, The known test beam parameters correspond to a preset fixed deflection angle or a preset scanning trajectory; the comparison of the measured feedback data with the pre-stored reference feedback data to obtain the model drift includes: Calculate the angular deviation between the measured far-field main lobe position and the reference main lobe position, as well as the difference between the measured sidelobe suppression ratio and the reference sidelobe suppression ratio, and quantitatively evaluate the output drift of the target phase prediction model. By comparing the deviation between the actual driving voltage and the theoretical voltage of each channel, the drift of the independent calibration mapping relationship is evaluated. The triggering of rapid recalibration of the independent calibration mapping relationship includes: re-performing a sine function fitting scan on channels whose drift exceeds a preset calibration update threshold, and updating their phase shift voltage values ​​and independent calibration mapping relationships.

4. The data transmission method according to claim 1, characterized in that, The independent calibration mapping relationship is established based on the following method: A varying driving signal is applied to the phase shifter of each channel in the optical phased array, and the periodic change in far-field diffraction efficiency is monitored. The phase shift voltage value of each channel is obtained independently by fitting the periodic change with a sine function; Based on the phase shift voltage value, a mapping between channel identifier and driving voltage is established.

5. The data transmission method according to claim 1, characterized in that, The target phase prediction model is a model trained using a loss function that includes direction vector constraints and an activation function that matches phase periodicity; wherein, the direction vector constraints are used to eliminate far-field confusion caused by phase translation.

6. A data transmission device based on an optical phased array, characterized in that, It includes a processor and a memory, the memory storing a computer program that, when executed by the processor, implements the data transmission method as described in any one of claims 1 to 5.

7. A data transmission system based on an optical phased array, characterized in that, It includes a signal transmitter, a signal receiver, and the data transmission device as described in claim 6; The signal transmitting end and the signal receiving end each include an optical phased array. The data transmission device is used to control the beam deflection of the optical phased array to realize data transmission between the signal transmitting end and the signal receiving end.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the data transmission method as described in any one of claims 1 to 5.

Images