Network communication method and system based on optical modules
By collecting and parsing network communication requests, a digital twin link is constructed for policy pre-simulation, which solves the problem of isolated upper and lower layer states in optical module network communication, realizes dynamic reconfiguration and resource scheduling of optical modules, and ensures the stability and availability of communication.
Patent Information
- Application Number
- CN202611300659.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-26
- Publication Date
- 2026-11-10
AI Technical Summary
In existing network communication technologies based on optical modules, upper-layer network communication requests and the physical layer status of the underlying optical modules are isolated, lacking joint perception of global information. This makes it impossible for optical modules to adjust their strategies according to business needs, which can easily lead to link interruptions or signal oscillations and affect business continuity.
The system collects network communication requests and analyzes communication demand indicators, synchronously collects multi-dimensional operating status parameters of optical modules, constructs a digital twin link for strategy pre-simulation, generates optical layer communication strategies, and monitors transmission quality in real time to adjust optical layer strategies, avoiding trial and error on real physical links.
It enables dynamic reconfiguration of optical modules and intelligent scheduling of multi-core optical fiber resources, preventing link interruptions and signal oscillations, and ensuring high availability of communication in highly sensitive business scenarios.
Smart Images

Figure CN123075307A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of network communication technology, and in particular to a network communication method and system based on optical modules. Background Technology
[0002] With the rapid development of supercomputing centers and large-scale distributed intelligent computing networks, data interaction between nodes presents stringent requirements for high bandwidth, low latency, and extremely high reliability. As the core physical layer hardware of optical communication, the operating status and scheduling strategy of optical modules directly determine the overall network performance. However, existing network communication technologies based on optical modules have the following shortcomings in practical applications: In existing technologies, the service intent of upper-layer network communication requests and the physical layer operating status of the underlying optical modules are isolated from each other. Optical modules often use static configuration or blind trial and error mechanisms for parameter tuning, lacking joint perception of global information. When the physical link deteriorates, existing optical modules cannot make targeted strategy adjustments based on actual service needs, and directly perform parameter trial and error on the real physical link. This can easily lead to link interruption or signal oscillation due to improper parameter configuration, seriously affecting service continuity. Summary of the Invention
[0003] The purpose of this invention is to overcome the shortcomings of the prior art. This invention provides a network communication method and system based on optical modules.
[0004] This invention provides a network communication method based on an optical module, comprising: Collect the corresponding network communication requests, determine the corresponding communication requirement indicators based on the parsing of the network communication requests, and simultaneously collect the multi-dimensional operating status parameters of the optical module under the current link. The multi-dimensional operating status parameters include the optical transceiver signal-to-noise ratio, offset point drift, and dynamic channel crosstalk value across multi-core optical fibers of the optical module. A digital twin link for the optical module is constructed based on communication demand indicators and multi-dimensional operating status parameters. In the digital twin link, the joint optimization objective is to maximize bandwidth utilization and minimize end-to-end latency. The strategy is pre-drilled for the joint optimization objective to generate the corresponding optical layer communication strategy. The optical layer communication strategy is sent to the silicon photonics integrated engine of the optical module, which drives the silicon photonics integrated engine to complete the electro-optical signal conversion according to the target modulation format and forward error correction code coding rate, and establishes the target optical communication path according to the mapping relationship of the spatial multiplexing channel. During the transmission of data streams through the target optical communication path, the transmission quality parameters of network communication are monitored in real time. When the transmission quality parameters are lower than the preset quality parameter threshold, the corresponding abnormal data is marked, and the optical layer communication strategy of the optical module is adjusted in reverse.
[0005] This invention provides a network communication system based on optical modules, which is applied to the aforementioned network communication method based on optical modules; the network communication system based on optical modules includes: The parameter module is used to collect the corresponding network communication requests, determine the corresponding communication requirement indicators based on the parsing of the network communication requests, and simultaneously collect the multi-dimensional operating status parameters of the optical module under the current link. The multi-dimensional operating status parameters include the optical transceiver signal-to-noise ratio, offset point drift, and dynamic channel crosstalk value across multi-core optical fibers of the optical module. The strategy module is used to construct a digital twin link of the optical module based on communication demand indicators and multi-dimensional operating status parameters. In the digital twin link, the joint optimization objective is to maximize bandwidth utilization and minimize end-to-end latency. The strategy pre-plays the joint optimization objective to generate the corresponding optical layer communication strategy. The communication path module is used to send the optical layer communication strategy to the silicon photonics integrated engine of the optical module, drive the silicon photonics integrated engine to complete the electro-optic signal conversion according to the target modulation format and forward error correction code coding rate, and establish the target optical communication path according to the mapping relationship of the spatial multiplexing channel. The adjustment module is used to monitor the transmission quality parameters of network communication in real time during the transmission of data streams through the target optical communication path. When the transmission quality parameters are lower than the preset quality parameter threshold, the corresponding abnormal data is marked, and the optical layer communication strategy of the optical module is adjusted in reverse.
[0006] Compared with the prior art, the beneficial effects of the present invention are: (1) Collect the corresponding network communication requests, determine the corresponding communication demand indicators based on the parsing of the network communication requests, and simultaneously collect the multi-dimensional operating status parameters of the optical module under the current link. The multi-dimensional operating status parameters include the optical transceiver signal-to-noise ratio, offset point drift, and dynamic channel crosstalk value across multi-core optical fibers of the optical module. Based on the communication demand indicators and multi-dimensional operating status parameters, construct a digital twin link of the optical module. In the digital twin link, maximize bandwidth utilization and minimize end-to-end latency as the joint optimization objectives. Perform strategy pre-playing on the joint optimization objectives to generate the corresponding optical layer communication strategy. This avoids trial and error on the real physical link and prevents link interruption or oscillation caused by improper parameter configuration.
[0007] (2) The optical layer communication strategy is sent to the silicon photonics integrated engine of the optical module, which drives the silicon photonics integrated engine to complete the electro-optic signal conversion according to the target modulation format and forward error correction code coding rate, and establishes the target optical communication path according to the mapping relationship of the space division multiplexing channel. This realizes the dynamic reconstruction of the underlying physical parameters and the intelligent scheduling of multi-core optical fiber space division resources, effectively avoids the deterioration of fiber cores, and maximizes the transmission potential of the multi-dimensional multiplexing architecture.
[0008] (3) During the transmission of data through the target optical communication path, the transmission quality parameters of the network communication are monitored in real time. When the transmission quality parameters are lower than the preset quality parameter threshold, the corresponding abnormal data is marked and the optical layer communication strategy of the optical module is adjusted in reverse. The physical link can be adaptively reconstructed before the actual packet loss of the service occurs, which greatly ensures the high availability of communication in highly sensitive service scenarios such as supercomputing. Attached Figure Description
[0009] Figure 1 This is a flowchart illustrating the network communication method based on optical modules in an embodiment of the present invention. Figure 2 This is a simplified schematic diagram illustrating a specific application scenario of the network communication method based on optical modules in this embodiment of the invention. Figure 3 This is a schematic diagram of the structural composition of a network communication system based on an optical module according to an embodiment of the present invention. Detailed Implementation
[0010] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.
[0011] Please see Figures 1 to 3 A network communication method based on optical modules, applied to network communication scenarios; the network communication method based on optical modules includes: Step S11: Collect the corresponding network communication request, determine the corresponding communication requirement index based on the parsing of the network communication request, and simultaneously collect the multi-dimensional operating status parameters of the optical module under the current link. The multi-dimensional operating status parameters include the optical transceiver signal-to-noise ratio, offset point drift, and dynamic channel crosstalk value across multi-core optical fibers of the optical module. Step S12: Construct a digital twin link for the optical module based on communication demand indicators and multi-dimensional operating status parameters. In the digital twin link, maximize bandwidth utilization and minimize end-to-end latency are the joint optimization objectives. Perform strategy pre-playing on the joint optimization objectives to generate the corresponding optical layer communication strategy. Step S13: Send the optical layer communication strategy to the silicon photonics integrated engine of the optical module, drive the silicon photonics integrated engine to complete the electro-optical signal conversion according to the target modulation format and forward error correction code coding rate, and establish the target optical communication path according to the mapping relationship of the spatial multiplexing channel. Step S14: During the transmission of data stream through the target optical communication path, monitor the transmission quality parameters of network communication in real time. When the transmission quality parameters are lower than the preset quality parameter threshold, mark the corresponding abnormal data and adjust the optical layer communication strategy of the optical module in reverse.
[0012] In step S11, the specific steps are as follows: S111: Based on the network communication requests collected by the data receiving end, perform deep protocol parsing on the network communication requests and extract communication requirement indicators at the business level. The communication requirement indicators include preset end-to-end latency thresholds, packet loss rate sensitivity levels, and bandwidth fluctuation ranges. S112: Through a miniature sensor array and high-speed sampling circuit deployed inside the optical module, multi-dimensional operating status parameters of the optical module under the current link are collected synchronously. Among them, the pilot signal detection unit integrated on the edge of the multi-core fiber is used to measure and calculate the dynamic channel crosstalk value between adjacent cores in the multi-core fiber in real time. Combined with the signal power statistics and bit error distribution characteristics of the optical transceiver, the optical signal-to-noise ratio is calculated. By monitoring the dynamic response curve of the DC bias voltage and optical output power of the silicon photonic modulator, the bias point drift is calibrated, and then a full-domain perception dataset including service intent and physical layer status is constructed.
[0013] In step S111, the specific steps are as follows: At this time, based on the data receiving end, the corresponding network communication request is collected. The data receiving end continuously bypasses and listens to or directly receives data streams from the upstream switching network or host bus, and captures Ethernet data frames, IP packets or Infiniband data packets as the original data source for the network communication request.
[0014] The network communication request is subjected to deep protocol parsing. This process is specifically manifested as follows: the hardware-accelerated parsing logic inside the data receiving end, such as FPGA logic array or ASIC chip, strips the packet header level by level according to the network protocol stack; identifies the link layer and network layer headers, and extracts VLAN tags and source / destination IP addresses to determine the flow forwarding path.
[0015] More importantly, it involves in-depth analysis of transport and application layer messages; for example, extracting port information and data segment length from the transport layer TCP / UDP header to determine whether the current traffic is characterized by long-term flow or short-term bursts; furthermore, it identifies and parses application layer-specific protocol markers, such as opcodes in the RoCEv2 (RDMA over Converged Ethernet) protocol header, service transport type in infinite band data packets, or synchronization communication primitive markers in the MPI (Message Passing Interface) protocol in high-performance computing. Through the above-mentioned layer-by-layer decapsulation and feature field extraction, it completes in-depth protocol analysis of network communication requests.
[0016] Extract communication requirement indicators at the business level; based on the field features obtained from the aforementioned deep protocol parsing, the data receiver maps the parsing results to a set of quantifiable communication requirement indicators at the business level, which includes at least a preset end-to-end latency threshold, packet loss rate sensitivity level, and bandwidth fluctuation range.
[0017] The extraction process for the preset end-to-end latency threshold is as follows: based on the parsed application layer protocol type or network layer DSCP (Differentiated Services Code Point) field, match it with the locally stored latency SLA (Service Level Agreement) mapping table; for example, when the parsed packet belongs to an RDMA Send / Write operation, it is determined that the service is extremely sensitive to latency, and the mapped and extracted end-to-end latency threshold is set in the sub-microsecond range, such as 0.5μs; if it is a normal TCP large file transmission stream, the mapped and extracted latency threshold can be relaxed to the millisecond range, such as 5ms.
[0018] The extraction process for the packet loss rate sensitivity level is as follows: based on the sequence number continuity check of the transport layer message, the confirmation and retransmission mechanism requirements of the application layer protocol, or by directly reading the packet loss sensitivity flag reserved in the message header, the packet loss rate sensitivity level is classified and labeled. For example, if the parsed data is distributed machine learning model parameter synchronization traffic, such as the parameter set All-Reduce operation, since its retransmission will cause the computing power of the computing node to idle, the extracted packet loss rate sensitivity level is labeled as "extremely high sensitivity, with an allowable bit error rate of less than 10^-12"; if it is ordinary video streaming media traffic, the extracted packet loss rate sensitivity level can be labeled as "medium-low sensitivity".
[0019] The extraction process for the bandwidth fluctuation range is as follows: Based on the transport layer window size, application layer message payload length, and data stream packet arrival time interval in the parsed message, the data receiver uses a sliding window statistical algorithm to calculate the average transmission rate and peak burst rate of the current service traffic, thereby determining the bandwidth fluctuation range; for example, if it is found that the message density of a certain storage traffic exhibits periodic abrupt changes within the window period, then the bandwidth fluctuation range is extracted as "100Gbps to 400Gbps high-frequency oscillation".
[0020] In step S112, the specific steps are as follows: At this time, through the miniature sensor array and high-speed sampling circuit deployed inside the optical module, the multi-dimensional operating status parameters of the optical module under the current link are collected synchronously. Among them, on the silicon photonic integrated chip substrate of the optical module, or on the high-frequency circuit board adjacent to the optical transceiver interface, a miniature sensor array consisting of a miniature thermal sensor, a high-frequency photodetector array, and a high-speed analog-to-digital converter (ADC) is solidified and integrated. When the optical signal is transmitted in the link, the high-speed sampling circuit discretizes the analog electrical signal and optical detection signal in the link at a preset "high sampling rate, such as gigahertz level", thereby synchronously acquiring multi-dimensional raw operating status parameters including optical power, level amplitude, phase characteristics, etc.
[0021] Optionally, for the micro-sensor array, which is not a general collection of sensors, but rather an integrated monitoring unit array directly fabricated on the silicon photonics chip substrate of the optical module using silicon photonics monolithic integration technology. Specifically, this micro-sensor array consists of three parts: The first part is an on-chip photodetector array based on the epitaxial growth of germanium-silicon material. Its photosensitive surface size is 10 micrometers by 10 micrometers, and its response wavelength covers 1,260 nanometers to 1,620 nanometers. It is deployed at the output end of the silicon photomodulator, the monitoring port of the waveguide coupler, and the front end of the multi-core fiber fan-in device to collect optical power and signal amplitude information. The second part is a thermal sensing unit based on a silicon-based microring resonator structure. This unit is located on the same chip layer as the optical waveguide. The local temperature of the chip is calculated by monitoring the thermal drift of the microring resonator wavelength, and the temperature measurement accuracy reaches ±0.1 degrees Celsius. The third part is a sampling circuit based on a high-speed analog-to-digital converter array. This circuit is also integrated into the chip edge area using silicon-based complementary metal-oxide-semiconductor technology. It has a sampling rate of 10 gigabits per second and a quantization accuracy of 8 bits, and can perform real-time discretization acquisition of the analog electrical signal output by the photodetector.
[0022] The three parts mentioned above are interconnected by a metal interconnect layer within the chip, and the entire miniature sensor array occupies a chip area of no more than two square millimeters, with power consumption controlled to within fifty milliwatts. For data synchronization, the high-speed sampling circuit uses a 250 MHz reference clock generated by the phase-locked loop inside the optical module as the synchronization base, performing parallel sampling of signals from each sensing unit at the same time. The sampled data is then simplified using the chip's built-in data compression logic before being output to the digital signal processor of the optical module. This ensures that the multi-dimensional operating state parameters are strictly aligned on the time axis, providing a hardware-level source of raw data with a unified time base for the subsequent construction of the global sensing dataset.
[0023] Furthermore, for the measurement and calculation of dynamic channel crosstalk values between adjacent cores in a multi-core fiber, the specific implementation method is as follows: in each independent core of the multi-core fiber (MCF), in addition to the main signal carrying data, a pilot signal detection unit integrated in the core edge (or cladding region) is used for monitoring.
[0024] Optionally, the pilot signal detection unit is not an abstract functional module, but a specific hardware structure implemented using fiber Bragg grating-assisted coupling technology at the coupling interface between the multi-core fiber fan-in / fan-out device and each individual fiber core. This pilot signal detection unit consists of a narrowband reflective fiber Bragg grating written in the cladding and inter-core region of each fiber core, and a corresponding on-chip wavelength demodulation circuit.
[0025] Specifically, fiber Bragg gratings with specific central reflection wavelengths are written at the input and output ends of each fiber core. These reflection wavelengths are at least 40 nanometers apart from the main data communication wavelength (such as the 1,550-nanometer band), thereby ensuring that the pilot signal and the data signal are completely separated in the frequency domain and avoiding mutual interference.
[0026] When it is necessary to measure the dynamic channel crosstalk value between adjacent fiber cores, the transmitter injects a continuous wave pilot light signal into the pilot fiber core through an independent low-frequency laser. The wavelength of this pilot light is precisely matched with the reflection wavelength of the corresponding fiber Bragg grating. When the pilot light signal leaks to an adjacent fiber core due to mode coupling caused by fiber microbending or temperature changes, the fiber Bragg grating on the end face of the adjacent fiber core reflects the leaked pilot light back into the optical path in front of the fan-in device. The wavelength selective coupler integrated at the front end of the fan-in device then extracts the light and sends it to the on-chip wavelength demodulation circuit for intensity detection.
[0027] The on-chip wavelength demodulation circuit employs a combination of an arrayed waveguide grating and a balanced photodetector, enabling it to detect changes in the optical power of leaked pilot light with nanowatt-level power resolution and convert them into a voltage signal proportional to the crosstalk intensity. The optical insertion loss of the entire pilot signal detection unit is less than 0.5 dB, and its impact on signal attenuation in the main data channel is negligible. Furthermore, all grating writing processes can be completed in a single step during the fabrication stage before the multi-core fiber leaves the factory, ensuring engineering feasibility.
[0028] In practice, the transmitter injects a pilot signal with a frequency or wavelength in the low-frequency out-of-band region into a specific fiber core. When the optical fiber is subjected to external physical micro-bending, twisting, or temperature gradient changes, mode coupling leakage occurs between fiber cores. At this time, the "pilot signal detection unit" integrated at the edge of adjacent fiber cores, such as a specific etched waveguide detector, will extract this leaked out-of-band pilot signal. By calculating the intensity change and phase shift of the pilot signal, and using a preset crosstalk transfer function model, the dynamic channel crosstalk value between adjacent fiber cores in the multi-core optical fiber is calculated in real time, expressed in decibels (dB).
[0029] Next, the specific implementation method for calculating the optical signal-to-noise ratio (SNR) is as follows: combining the signal power statistics collected by the aforementioned micro-sensor array at the optical transceiver end with the bit error distribution characteristics fed back by the underlying digital signal processing (DSP) unit for joint estimation; wherein, the average optical received power of the current signal is obtained through the high-speed photodiode at the optical receiver end, and the noise power is obtained through filtering and measuring the link noise floor, thereby obtaining the preliminary carrier-to-noise ratio.
[0030] Meanwhile, the forward error correction (FEC) decoder in the DSP is used to statistically analyze the bit error rate before decoding and its variance distribution characteristics within the time window in real time. This bit error distribution characteristic is used as an actual reflection of channel degradation and substituted into the preset SNR-BER empirical mapping model to perform nonlinear correction on the initial carrier-to-noise ratio, thereby accurately calculating the optical transceiver signal-to-noise ratio under the current complex modulation format.
[0031] The specific implementation method for calibrating the bias point drift of a silicon photonic modulator is as follows: The transmission characteristic curve of the silicon photonic modulator (e.g., a Mach-Zehnder modulator, MZM) inside the optical module will shift due to the thermo-optical effect and thermal expansion and contraction of silicon material. Therefore, the DC bias voltage applied to the silicon photonic modulator is monitored in real time using a high-speed sampling circuit. Simultaneously, the optical output power is collected using a monitoring photodetector (MPD) located at the modulator output. By superimposing a weak low-frequency jitter signal onto the DC bias voltage, the response amplitude and phase of the output optical power are demodulated, and a dynamic response curve of the DC bias voltage and optical output power is plotted. The current dynamic response curve is compared with the ideal sinusoidal response curve calibrated at the factory, and the difference in the horizontal axis voltage at the operating point of the two curves is extracted. This difference is calibrated as the current bias point drift of the silicon photonic modulator.
[0032] Optionally, for the weak low-frequency jitter signal superimposed on the DC bias voltage, its frequency, amplitude, and waveform type are subject to clear and operable quantitative constraints. Regarding frequency, the frequency of this jitter signal is set far below the lower limit of the high-speed data signal bandwidth, specifically between 10 kHz and 100 kHz, to prevent spectral overlap with high-speed modulation signals at the gigahertz level. Simultaneously, this frequency must be higher than the reciprocal of the thermal time constant of the silicon photonic modulator to ensure real-time tracking of the slow-varying characteristics caused by thermal drift. Regarding amplitude, the peak-to-peak amplitude of this jitter signal is strictly limited to within 5% of the DC bias voltage amplitude. The specific value is proportionally calculated based on the nominal value of the current modulator half-wave voltage to ensure that the optical output power fluctuation caused by the superposition of this jitter signal does not exceed ±2% of the modulator's average output optical power, thereby avoiding the introduction of additional nonlinear distortion or significant degradation of the modulation quality of the main signal.
[0033] Regarding waveform type, the jitter signal adopts a sine wave form to facilitate subsequent phase-sensitive detection via a lock-in amplifier, thereby accurately extracting the fundamental component and phase information of the output optical power response. Under the above parameter constraints, the system acquires the output optical power response curve by monitoring the photodetector, and inputs the fundamental amplitude and phase information of the response curve, which are at the same frequency as the jitter signal, into a preset transmission characteristic model. After demodulation, the offset voltage value of the current bias point relative to the ideal quadrature operating point can be quantized, thus completing the accurate calibration of the bias point drift.
[0034] Based on the above extraction process, the "communication requirements indicators: end-to-end latency threshold, packet loss rate sensitivity level and bandwidth fluctuation range" parsed in step S111 above are strictly aligned and fused with the "physical layer multi-dimensional operating status parameters: dynamic channel crosstalk value, optical transceiver signal-to-noise ratio, and offset point drift" collected and calculated in step S112 above. This results in the construction of a full-domain perception dataset that includes service intent and physical layer status. This full-domain perception dataset is stored in the local cache of the optical module in the form of a structured matrix or tensor, providing a complete input source for the subsequent construction of the digital twin link.
[0035] Optionally, the construction process of the full-domain perception dataset is not simply a matter of splicing business requirements with physical parameters, but rather a process that generates structured data units through strict timestamp alignment and feature-level fusion algorithms. Specifically, the system uses the high-speed clock signal inside the optical module as a reference to unify the communication requirement indicators parsed in step S111 and the multi-dimensional operating status parameters collected in step S112 to the same millisecond-level time coordinate.
[0036] During the fusion process, the physical layer parameters are first preprocessed: the optical transmit / receive signal-to-noise ratio is converted into a signal-to-noise ratio degradation factor in the complex domain, the offset point drift is mapped to the phase offset angle of the Mach-Zehnder modulator transmission curve, and the dynamic channel crosstalk value across multi-core optical fibers is converted into a mode coupling coefficient matrix between adjacent fiber cores.
[0037] The system employs a dimensionality reduction and fusion algorithm based on principal component analysis to weightedly integrate the transformed physical layer features with the end-to-end latency threshold, packet loss rate sensitivity level, and bandwidth fluctuation range of the service layer, generating a feature tensor with unified dimensions and normalized units. This feature tensor contains three dimensions: timestamp labels, service intent constraint vectors, and physical channel feature vectors. Following a circular buffer storage structure, it is rolled over and stored in the local cache of the optical module in units of preset-length time windows, thus simultaneously serving as a data source for digital twin link construction and subsequent optimization simulations. The dataset also incorporates a missing value handling mechanism. When physical parameters at a certain sampling moment fail to be successfully acquired due to transient interference, the system uses a linear prediction extrapolation method based on the measurement value at the previous moment to fill in the missing data, ensuring the continuous and complete data flow input to the digital twin space, thereby guaranteeing the stability and reliability of the twin model's operation.
[0038] In step S12, the specific steps are as follows: S121: Input communication demand indicators and multi-dimensional operating status parameters into the same digital twin space along different paths, and perform multi-level fusion in the digital twin space to gradually build the digital twin link of the optical module. The digital twin link can map the nonlinear effects and multi-core crosstalk environment of the current optical channel in real time. S122: A joint optimization mechanism is introduced into the digital twin link, with the joint optimization objectives of maximizing bandwidth utilization and minimizing end-to-end latency, and link stability as a constraint condition; multiple rounds of strategy pre-playing are performed on the joint optimization objectives, traversing different modulation formats, forward error correction coding rates and power configuration combinations in the digital twin space, thereby outputting the optimal solution that meets the preset convergence conditions, and using it as the optical layer communication strategy. This optical layer communication strategy includes an adaptive resource configuration table for different multi-core fiber cores and the corresponding set of signal processing parameters.
[0039] In step S121, the specific steps are as follows: At this point, communication requirement indicators and multi-dimensional operational status parameters are input to the same digital twin space along different paths. To eliminate the heterogeneity between service layer data and physical layer data in terms of acquisition timing, data frequency, and dimensions, the microcontroller inside the optical module adopts a physically isolated dual-channel data transmission bus. The "communication requirement indicators" at the service layer (end-to-end latency threshold, packet loss rate sensitivity level, and bandwidth fluctuation range) are used as logical constraints and input to the top constraint layer of the digital twin space along the low-frequency control bus. Meanwhile, the "multi-dimensional operational status parameters" of the optical module under the current link (dynamic channel crosstalk value, optical transmit / receive signal-to-noise ratio, and offset point drift) are used as physical channel characteristics and input to the bottom physical modeling layer of the digital twin space along the high-speed data bus. This dual-path independent input mechanism avoids the parsing errors caused by the aliasing of high and low frequency signals and ensures that the twin space can simultaneously receive the intended instructions and the physical reality.
[0040] Multi-level fusion is performed in the digital twin space to gradually construct the digital twin link of the optical module. This multi-level fusion process is specifically divided into two stages: feature-level fusion and model-level fusion. In the feature-level fusion stage, after receiving the multi-dimensional operating status parameters of the physical layer, the digital twin space performs coordinate system transformation and timestamp alignment on the parameters. Specifically, the scalar form of optical transceiver signal-to-noise ratio is converted into a complex domain signal-to-noise ratio degradation factor; the offset point drift is mapped to the phase offset parameter of the Mach-Zehnder modulator transmission curve; and the bandwidth fluctuation range of the service layer is converted into the number of frequency domain resource blocks to be scheduled, forming a unified feature vector matrix.
[0041] The built-in modeling engine of the digital twin space calls upon pre-set optical physical equations and network topology constraint models. Specifically, using the aforementioned eigenvector matrix, it drives a pre-set nonlinear Schrödinger equation (NLS) solver, generating a virtual optical fiber channel transmission model with the actual multi-core fiber structure of the current link as boundary conditions. Simultaneously, it uses the offset point drift to drive the transfer function model of the silicon photonic modulator, generating a virtual electro-optical conversion front-end model. The end-to-end delay threshold and packet loss sensitivity level of the service layer in the top-level constraint layer are used as boundary penalty functions and injected into the iterative process of solving the physical equations, achieving deep coupling between the intent and the physical model, thereby gradually constructing a high-fidelity optical module digital twin link.
[0042] This enables the digital twin link to map the nonlinear effects and multi-core crosstalk environment of the current optical channel in real time. This feature is achieved by embedding a real-time evolution mechanism in the twin link. Specifically, the high-frequency data receiving port inside the digital twin link continuously monitors the fluctuations of multi-dimensional operating status parameters on the high-speed data bus. When the dynamic channel crosstalk value or the optical transmit / receive signal-to-noise ratio changes slightly, the nonlinear effect calculation module in the twin model synchronously updates the intensity coefficients of the Kerr effect and Raman scattering effect.
[0043] Optionally, the digital twin link is not a simple software mirror of the physical optical module, but a virtual simulation platform with rigorous mathematical mapping relationships and a real-time data synchronization mechanism. Specifically, it is implemented by pre-setting a nonlinear Schrödinger equation solver and a silicon photonics modulator transfer function model in the local edge computing unit of the optical module to form the basic physical modeling engine. During operation, the system uses dynamic channel crosstalk values, optical transmit / receive signal-to-noise ratio, and bias point drift acquired through the high-speed data bus as boundary conditions and initial parameters, which are then input into this physical modeling engine.
[0044] Based on the above parameters, the engine uses a step-by-step Fourier method to iteratively solve the transmission equation of the optical signal in the multi-core optical fiber. Simultaneously, it combines the offset point drift to drive the modulator transfer function, reconstructing in real-time in virtual space an optical channel model with the same dispersion, nonlinearity, and crosstalk characteristics as the physical link. To ensure the fidelity between the virtual model and the physical entity, the system incorporates a residual monitoring mechanism. Every preset microsecond time window, the predicted receiver signal quality parameters of the digital twin link are compared with the actual monitored values fed back by the physical link. A Kalman filter algorithm is then used to perform closed-loop correction on the Kerr effect coefficients and mode coupling matrix within the model. This allows the digital twin link to accurately map the real-time evolution state of the physical optical channel with sub-millisecond delays, providing a highly reliable virtual testing environment for subsequent strategy simulations.
[0045] Meanwhile, the multi-core crosstalk environment calculation module synchronously updates the mode coupling coefficient matrix between adjacent fiber cores; thus, the digital twin link can accurately reproduce the dynamic physical evolution of dispersion, self-phase modulation, cross-phase modulation and multi-core crosstalk that is happening in the physical world in the optical fiber with sub-millisecond latency in the virtual space, providing a "sandbox" testbed that is highly consistent with the real physical environment for subsequent strategy pre-simulation.
[0046] In step S122, the specific steps are as follows: At this point, a joint optimization mechanism is introduced into the digital twin link, with the joint optimization objectives of maximizing bandwidth utilization and minimizing end-to-end delay, and link stability as a constraint. This joint optimization mechanism constructs a multi-objective cost function within the digital twin space of the optical module. In this cost function, the first objective function F1 is defined as maximizing the bandwidth utilization of spectrum resources and the spatial multiplexing channel, and its value monotonically increases with the increase of the spectrum allocation width. The second objective function F2 is defined as minimizing the end-to-end delay, which includes the physical transmission delay caused by group velocity dispersion of the optical signal in the multi-core fiber, and the processing delay introduced by forward error correction coding and decoding, and its value increases with the decrease of modulation order and FEC coding rate. The above two objective functions are weighted and summed to form the joint optimization objective function.
[0047] Meanwhile, link stability is taken as a hard constraint on the joint optimization objective function. Specifically, the packet loss rate sensitivity level extracted in step S111 is converted into the upper limit threshold of the bit error rate before forward error correction decoding, and the nonlinear effects mapped in the twin link in step S121 and the noise power generated by the multi-core crosstalk environment are restricted to the range that makes the lower limit of the signal-to-noise ratio meet the bit error rate threshold. Only in the solution space that meets the above constraints can the joint optimization mechanism allow subsequent optimization calculations.
[0048] The joint optimization objective is subjected to multiple rounds of strategy pre-playing, traversing different modulation formats, forward error correction coding rates, and power configuration combinations in a digital twin space, thereby outputting the optimal solution that satisfies the preset convergence conditions, which serves as the optical layer communication strategy. The digital twin space incorporates heuristic solution algorithms, such as a non-dominated sorting genetic algorithm with an elite retention strategy or a deep reinforcement learning agent. During the pre-playing process, the algorithm traverses multiple rounds in a preset discrete action space, which includes: "different modulation formats, such as quadrature phase shift keying (QPSK), 16th-order quadrature amplitude modulation (16QAM), and 64QAM"; "different forward error correction coding rates, such as coding rates of 0.8, 0.85, and 0.9"; and "different power configuration combinations, such as the bias current level of the transmitter's adjustable semiconductor optical amplifier or the driving voltage swing of the MZM modulator."
[0049] Regarding the preset convergence condition, it is not simply described as the fitness value increment being less than a certain undetermined threshold. Instead, it is clearly and quantitatively defined using a dual-criteria joint determination method based on the real-time requirements and accuracy requirements of different business scenarios. The first criterion is the relative improvement rate of the objective function value. That is, after each round of iterative strategy pre-playing, the system calculates the percentage relative deviation between the optimal fitness value obtained in the current round and the optimal fitness value obtained in the previous round. When this relative deviation percentage is lower than the first preset percentage threshold for three consecutive rounds, the optimization process is determined to have entered the saturation region.
[0050] For services with different latency sensitivity levels, the first preset percentage threshold is set differently: for extremely low latency services with an end-to-end latency threshold of less than ten microseconds, the relative improvement rate threshold is set to 0.5% to ensure rapid convergence within a limited number of iterations to meet strict latency constraints; for ordinary latency-tolerant services, the threshold is relaxed to one percent to reduce computational overhead.
[0051] The second criterion is the absolute upper limit of iteration rounds, which is the maximum allowed number of strategy pre-runs in the digital twin space. This maximum number is dynamically and adaptively adjusted based on the optical module control command issuance cycle and the service data stream arrival interval, and its value is limited to between fifty and three hundred times. This ensures the accuracy of optimization while preventing the optimization process from endlessly consuming computing resources. Only when either the relative improvement rate continuously reaches a preset percentage threshold or the number of iteration rounds reaches the preset maximum upper limit is the system determined to have met the preset convergence condition, terminates the strategy pre-run, and outputs the current optimal solution.
[0052] In each iteration, the digital twin link performs numerical deduction based on the currently selected action combination using internal optical physics equations to calculate the predicted bandwidth utilization, predicted end-to-end latency, and predicted bit error rate. These values are then substituted into the aforementioned joint optimization objective function to calculate the fitness value. When the number of iterations reaches a preset upper limit, or when the increment of the fitness value calculated in multiple consecutive iterations is less than a preset "convergence threshold," i.e., the preset convergence condition is met, the algorithm terminates the iteration and outputs the optimal solution on the Pareto front. This optimal solution is then encapsulated and defined as the optical layer communication strategy.
[0053] The optical layer communication strategy is defined as including an adaptive resource configuration table for different multi-core optical fibers and a corresponding set of signal processing parameters. The output optical layer communication strategy is structured into a two-dimensional data entity. The first dimension is the adaptive resource configuration table for different multi-core optical fibers. This table is indexed by the number of each independent fiber in the multi-core optical fiber and allocates the wavelength resources carried by the fiber in the current service cycle, the modulation format selection, and whether the fiber is activated as a primary or backup route. Since the crosstalk environment of each fiber is different, the resource configuration table allows different fibers to adopt heterogeneous transmission configurations.
[0054] The second dimension is the corresponding set of signal processing parameters, which is associated with the resource configuration table and includes parameters such as the target DC bias voltage compensation value of the silicon photonic modulator at the transmitter, the symbol rate of the driving signal, and the tap coefficient of the electrical equalizer in the digital signal processing (DSP) module at the receiver. The data sets of the above two dimensions together constitute a complete optical layer communication strategy that can be directly sent to the underlying hardware for execution.
[0055] In step S13, the specific steps are as follows: S131: The optical layer communication strategy is encapsulated into a corresponding control command stream and sent to the silicon photonics integrated engine of the optical module through a high-speed control interface; the silicon photonics integrated engine parses the control command stream, dynamically configures its internal digital-to-analog converter and drive circuit, thereby adjusting the amplitude and phase of the drive signal, and then loads the constellation map mapping according to the target modulation format, and completes the corresponding electro-optic signal conversion in conjunction with the target forward error correction code coding rate. S132: Obtain the mapping relationship of the spatial division multiplexing channel, and control the optical switch matrix and multi-core fiber fan-in and fan-out devices based on the mapping relationship to couple the loaded optical signal to a specific core of the target multi-core fiber, and establish the corresponding target optical communication path by combining the wavelength division multiplexing mechanism and the multi-dimensional multiplexing architecture.
[0056] In step S131, the specific steps are as follows: At this point, the optical layer communication strategy is encapsulated into a corresponding control command stream and sent to the silicon photonics integrated engine of the optical module through a high-speed control interface; wherein, the microcontroller or edge computing unit inside the optical module performs hardware-level data encapsulation of the "optical layer communication strategy, i.e., adaptive resource configuration table and signal processing parameter set" output in the aforementioned step S122.
[0057] During the packaging process, according to the register mapping specification or direct memory access (DMA) protocol supported by the silicon photonics integrated engine of the optical module, the logic-level strategy parameters are converted into binary or hexadecimal control instruction streams with explicit physical address pointers. In these control instruction streams, specific data segments correspond to the target modulation format code, the target forward error correction code rate value, the target gain coefficient of the digital-to-analog converter (DAC), and the bias voltage compensation value, respectively. Through the high-speed control interface inside the optical module, such as a parallel bus with a low number of pins and support for gigahertz-level clock synchronization, or a high-speed serial peripheral interface, the above control instruction stream is sent to the internal control register group of the silicon photonics integrated engine in DMA burst mode without interruption, thereby achieving instantaneous anchoring of configuration parameters.
[0058] The silicon photonics integrated engine parses the control command stream and dynamically configures its internal digital-to-analog converter and drive circuit to adjust the amplitude and phase of the drive signal. The control logic core integrated inside the silicon photonics integrated engine receives and decodes the control command stream, extracts the parameter fields, and writes them into the control register of the corresponding digital-to-analog converter (DAC) and the configuration port of the programmable gain amplifier (PGA).
[0059] In this stage, for amplitude adjustment, the digital-to-analog converter dynamically changes the swing of its output analog differential current or voltage according to the received digital amplitude control word. This analog signal is amplified by the drive circuit and then applied to the RF electrode of the silicon photonic modulator as an RF drive signal. For phase adjustment, the silicon photonic integrated engine changes the DC bias voltage difference applied to the two arms of the Mach-Zehnder interferometer through the internally integrated phase shifter control sub-circuit, thereby precisely controlling the phase difference between the two arms of optical carriers. This ensures that the amplitude and phase of the interferometric optical signal can accurately match the vector state required by the target modulation format. Through the above-mentioned coordinated dynamic configuration of the digital-to-analog converter and the drive circuit, the shaping of the physical characteristics of the transmitter's electrical domain signal is completed.
[0060] The constellation map is loaded according to the target modulation format, and the corresponding electro-optic signal conversion is completed in conjunction with the target forward error correction code coding rate. After the aforementioned physical layer simulation parameters are configured, the digital signal processing (DSP) logic unit inside the silicon photonics integrated engine begins to execute the algorithm-level mapping operation. The input high-speed binary data stream is sent to the forward error correction code (FEC) encoder. The encoder dynamically adjusts the generator matrix and parity bit insertion ratio according to the target FEC coding rate specified in the control command stream, and outputs the redundant bit stream after channel coding.
[0061] The redundant bit stream enters the constellation map mapping unit. According to the target modulation format specified in the control command stream, such as 16QAM or QPSK, the mapping unit takes "a number of consecutive bits, for example, 4 bits per symbol in 16QAM" as a symbol, looks up the constellation map lookup table stored locally, maps it to the ideal coordinate point on the complex plane, and generates the corresponding in-phase (I) and quadrature (Q) baseband signals. After the I / Q baseband signals are converted from digital to analog and driven and amplified, they are loaded onto the in-phase and quadrature dual-arm electrodes of the silicon photonic modulator to jointly modulate the amplitude and phase of the carrier optical signal, thereby completing the conversion from electrical digital signal to optical domain analog signal, that is, electro-optic signal.
[0062] In step S132, the specific steps are as follows: At this point, in step S132, the mapping relationship of the spatial multiplexing channel is obtained. The optical module internally stores the current topology table and channel mapping dictionary of the multi-core fiber (MCF). This mapping relationship not only records the correspondence between the "physical fiber core number, such as fiber core 1 to fiber core 7" and the logical channel identifier, but also records in real time the current dynamic channel crosstalk value and optical transceiver signal-to-noise ratio of each fiber core perceived based on the aforementioned steps S111 and S112. During the policy execution phase, the "adaptive resource configuration table" contained in the optical layer communication policy output by the digital twin link is extracted. The microcontroller inside the optical module consults this table to obtain which specific physical fiber cores the current service flow should be routed to, thereby establishing the target routing mapping relationship of the spatial multiplexing dimension.
[0063] Based on this mapping relationship, the optical switch matrix and the fan-in / fan-out devices of the multi-core fiber are controlled to couple the loaded optical signal to a specific core of the target multi-core fiber. The silicon photonics integrated engine of the optical module integrates a reconfigurable optical switch matrix. According to the aforementioned mapping relationship, the silicon photonics integrated engine applies a specific combination of bias voltages to the control electrodes of the optical switch matrix to change the refractive index of the waveguide inside the optical switch, thereby changing the on / off state of the optical signal and guiding the modulated optical signal to the output port corresponding to the target core.
[0064] The optical switch matrix is not a general optical path switching device, but a cross-switching network composed of thermo-optical or electro-optic silicon-based optical switch units, arranged in a two-dimensional array on a silicon photonic integrated chip.
[0065] Each basic switching unit consists of two three-port multimode interference couplers and two phase-shifting arms connecting the two couplers. One of the phase-shifting arms is covered with a metal heating electrode to achieve thermo-optical phase modulation, or a carrier injection type electro-optic phase modulator is used to achieve a microsecond-level response speed. The number of input ports of the optical switching matrix matches the number of output waveguides of the silicon photonic modulator, and the number of output ports corresponds to the total number of cores in the multi-core optical fiber, forming a complete input-output cross-connection topology.
[0066] In terms of control, based on the spatial multiplexing channel mapping relationship obtained in step S132, the system applies a DC voltage of preset amplitude to the heating electrode of the phase shift arm of the selected switching unit through a digital-to-analog converter. This voltage causes a 180-degree phase difference in the phase shift arm, thereby changing the output optical power distribution ratio of the Mach-Zehnder interferometer and guiding the input optical signal to the target output waveguide. For applications requiring fast switching, the system additionally configures an electronic switch array to provide a bias current injection mode, reducing the switching response time of the optical switch from the hundreds of microseconds level of the thermo-optic effect to the tens of nanoseconds level of the electro-optic effect. All switching units are addressed and driven through the chip's internal digital control register group, and control commands are simultaneously loaded to each switching unit via a parallel bus, ensuring that the entire optical switch matrix completes configuration updates within microseconds. The typical insertion loss of the optical switch matrix is controlled within 3 dB, and the inter-channel crosstalk is below -35 dB, which is sufficient to meet the signal routing requirements in multi-core fiber spatial multiplexing scenarios.
[0067] The optical signal enters a multi-core fiber fan-in / fan-out device, which is typically a three-dimensional waveguide integrated lens array or a special graded refractive index lens array. The fan-in / fan-out device utilizes a precise spatial optical coupling structure to accurately align and focus the independent optical signals from different output waveguides of the silicon photonic chip in physical space, so that they are coupled to the corresponding specific fiber core cross-sections on the end face of the multi-core fiber, thereby realizing the physical injection of spatial multiplexing.
[0068] For multi-core fiber fan-in and fan-out devices, the multi-core fiber fan-in and fan-out devices are not general coupling structures, but fan-in and fan-out couplers based on three-dimensional waveguide integration technology. The core of the coupler is a set of gradient refractive index lens arrays that have been precisely ground and polished, which form a coaxial optical alignment with the output waveguide array of the optical module.
[0069] The input end face of the fan-in fan-out device and the end face of the output waveguide of the silicon photonic chip adopt a combination of passive alignment and active fine-tuning assembly process: First, the waveguide array and the lens array are coarsely aligned at the sub-ten-micron level by the alignment marks on the silicon substrate. Then, a reference light signal is injected into the coupling optical path, and the lens array is finely adjusted in six degrees of freedom by the fine-tuning platform driven by the microelectromechanical system until the coupling efficiency reaches the maximum value. Finally, it is permanently fixed by UV curing adhesive or laser welding process.
[0070] In terms of structural parameters, the pitch length of the graded-index lens array is designed to be one-quarter pitch, so that the parallel light incident at the front end of the lens is focused at the output end into a spot with a diameter matching the diameter of the mode field of each core of the multi-core fiber, typically between nine and eleven micrometers. The output end face of the fan-in fan-out device is physically connected to the end face of the multi-core fiber through a high-precision ceramic ferrule and sleeve, with the connection gap controlled within ten micrometers. A refractive index matching fluid is filled at the connection interface to eliminate Fresnel reflection loss.
[0071] The total coupling loss of the entire fan-in and fan-out device is controlled within 1.5 dB, and the difference in coupling loss between channels does not exceed ±0.3 dB. This ensures that the signal injection power of each fiber core in the multi-core fiber is relatively balanced. Furthermore, the device operates in a temperature range of -20°C to 85°C, which is sufficient to meet the temperature fluctuation requirements of the supercomputing center computer room and the external interconnection environment.
[0072] Furthermore, a corresponding target optical communication path is established by combining wavelength division multiplexing (WDM) mechanism and multidimensional multiplexing architecture. Before or after coupling the optical signal to a specific fiber core, the optical module synchronously activates the WDM mechanism, and combines or demultiplexes optical carriers of different wavelengths in the frequency domain through the arrayed waveguide grating (AWG) or micro-ring resonator filter array integrated inside the silicon photonics integrated engine. The multidimensional multiplexing architecture is manifested here as a joint application system of "space division multiplexing: SDM, different fiber cores", "wavelength division multiplexing: WDM, different wavelengths" and "polarization multiplexing: PDM, orthogonal polarization state". When establishing the target optical communication path, the system combines and binds the physical route corresponding to the aforementioned specific fiber core, the allocated target wavelength channel and polarization state to form an end-to-end optical communication path with high-dimensional parallel capability and avoiding high crosstalk and high nonlinear degradation regions, thereby completing the reconstruction and connection of the entire physical channel.
[0073] In step S14, the specific steps are as follows: S141: During the transmission of data stream through the target optical communication path, the performance monitoring module integrated inside the optical module is used to perform real-time in-band detection of the signal and obtain the transmission quality parameters of the network communication. These transmission quality parameters include the real-time estimated value of the optical signal-to-noise ratio, the magnitude of the error vector, and the bit error rate statistics before soft-decision forward error correction decoding. S142: The transmission quality parameters are compared with the preset quality parameter threshold in real time. When the transmission quality parameters are lower than the preset quality parameter threshold, the anomaly handling mechanism is activated. A specific degradation marker is inserted into the control channel to mark the corresponding abnormal data, and the abnormal data is cached for subsequent retransmission or error correction. At the same time, closed-loop feedback control is triggered to feed back the abnormal status corresponding to the abnormal data to the digital twin link, and the modulation order, transmit power and routing mapping relationship in the optical layer communication strategy are adjusted in reverse.
[0074] In step S141, the specific steps are as follows: At this time, during the transmission of data through the target optical communication path, the performance monitoring module integrated inside the optical module performs real-time in-band detection of the signal. When the optical signal continuously transmits service data in the target fiber core spanning multiple optical fibers, the receiving end of the optical module does not interrupt or bypass the main data channel. Instead, it adopts in-band monitoring technology. This performance monitoring module is physically deployed inside the silicon photonics integrated receiving engine of the optical module. Specifically, it includes a miniature photodetector array coupled to the main optical signal splitter branch, and a high-speed digital signal processing (DSP) logic unit connected to the back end of the detector. By using an optical splitter, such as a directional coupler with extremely low insertion loss, it extracts weak optical power from the main optical signal at a preset ratio, such as 1% to 5%, as a monitoring signal and converts it into an electrical signal input to the DSP. Thus, without interrupting or changing the normal service data flow transmission, it achieves real-time and continuous in-band detection of the physical state of the target optical communication path that is currently in operation.
[0075] The transmission quality parameters of network communication are obtained, including the real-time estimate of the optical signal-to-noise ratio, the error vector magnitude, and the bit error rate statistics before soft-decision forward error correction decoding. The DSP unit within the aforementioned performance monitoring module calculates these three core transmission quality parameters based on the extracted monitoring signals using specific digital signal processing algorithms. Firstly, for the real-time estimation of optical signal-to-noise ratio (OSNR), the spectrum analysis module inside the DSP performs a fast Fourier transform (FFT) on the monitored signal, extracts the peak optical power at the center wavelength of the signal and the out-of-band noise floor power between adjacent channels, calculates the ratio of the two and deducts the splitting loss, and obtains the real-time estimation of the current optical signal-to-noise ratio.
[0076] Secondly, regarding the error vector magnitude (EVM), the DSP completes clock recovery and carrier phase synchronization, extracts the actual symbol coordinates received, compares them with the reference symbol coordinates on the ideal constellation diagram, calculates the root mean square value of the Euclidean distance between the two in the complex plane, and normalizes this root mean square error into the error vector magnitude, which is used to quantitatively evaluate the degree of constellation diagram divergence and distortion affected by nonlinear effects and crosstalk of the current modulation signal.
[0077] Third, regarding the bit error rate statistics before soft-decision forward error correction decoding, the DSP internally sets a statistical counter before the soft-decision FEC decoder. After equalization and decision of each data symbol segment, the hard-decision output bit stream is compared bit by bit with the known training sequence or CRC check code at the transmitting end, and the ratio of the number of erroneous bits to the total number of received bits is calculated to obtain the Pre-FEC bit error rate. This parameter does not depend on the result after FEC error correction and can most directly and proactively reflect the actual physical degradation of the underlying optical channel.
[0078] In step S142, the specific steps are as follows: At this point, the transmission quality parameters are compared with preset quality parameter thresholds in real time. When the transmission quality parameters fall below the preset quality parameter thresholds, an anomaly handling mechanism is activated. The microcontroller or programmable logic array inside the optical module pre-stores safety baseline thresholds for each transmission quality parameter. This threshold system is directly mapped from the packet loss rate sensitivity level extracted in step S111. For example, for "extremely sensitive" services, the preset Pre-FEC bit error rate threshold is set to 1×10⁻⁶. -3 The error vector amplitude (EVM) threshold is set to 12%, and the optical signal-to-noise ratio (OSNR) threshold is set to 23dB. The real-time comparison logic unit continuously receives the real-time transmission quality parameters output in step S141 at a clock cycle of nanoseconds and compares them with the above-mentioned preset thresholds by values greater than or less than the threshold. When any parameter exceeds the limit, for example, the real-time OSNR drops to 22dB, which is lower than the threshold of 23dB, the comparison logic unit immediately flips to the valid state and outputs a hardware interrupt signal, thereby triggering and starting the abnormal handling mechanism of the optical module.
[0079] Optionally, the preset quality parameter thresholds are not a set of fixed static values, but rather a dynamic threshold system jointly determined based on the service packet loss rate sensitivity level parsed in step S111 and the actual modulation format and coding rate of the current link. Specifically, the system internally stores a quality parameter baseline mapping table covering multiple modulation formats and forward error correction coding rate combinations. This mapping table is pre-calibrated based on the theoretical model of a standard optical communication system and offline simulation calibration data. For the three modulation formats—quadrature phase shift keying, 16th-order quadrature amplitude modulation, and 64th-order quadrature amplitude modulation—it provides safe boundary reference values for three parameters before forward error correction decoding: bit error rate, error vector amplitude, and optical signal-to-noise ratio.
[0080] During operation, the system first queries the mapping table to obtain the baseline threshold based on the target modulation format and coding rate combination selected in the current optical layer communication strategy. Then, based on the packet loss rate sensitivity level parsed in step S111, different security factors are applied to this baseline threshold: when the packet loss rate sensitivity level is extremely high, the system reduces the bit error rate threshold before forward error correction decoding by an order of magnitude from the baseline, reduces the error vector amplitude threshold by five percentage points, and increases the optical signal-to-noise ratio threshold by two dB; when the packet loss rate sensitivity level is medium to low, the baseline threshold is directly used without additional tightening. Furthermore, the system introduces a slow time-varying adaptive correction mechanism, which, at preset second-level time windows, fine-tunes the current threshold by incrementally increasing or decreasing it in small steps based on the accumulated bit error statistics and the number of automatic retransmission requests during actual transmission. This ensures service reliability while avoiding unnecessary policy adjustments triggered by overly strict threshold settings.
[0081] A specific degradation marker is inserted into the control channel to mark the corresponding abnormal data, and the abnormal data is buffered for subsequent retransmission or error correction. This process is performed at the junction of the link layer and physical layer of the optical module. When the abnormality handling mechanism is activated, the transmitting or receiving logic of the optical module inserts a specific degradation marker bit sequence in real time into the out-of-band control channel accompanying the service data stream. This degradation marker is used to notify the peer of the link or the upper-layer switch that the current data frame is in a physical layer degradation transmission state.
[0082] Simultaneously, the high-speed first-in-first-out (FIFO) memory or dedicated data buffer inside the optical module is activated. The control logic intercepts and buffers the "abnormal data streams: or data frames that failed the underlying verification" received during the time period corresponding to the degradation mark in the local memory. Through this operation, even if the physical link experiences a transient severe deterioration, these marked abnormal data are still retained locally in the optical module. After the physical link is repaired in subsequent steps, the underlying FEC's multiple retransmission mechanism or the retransmission protocol negotiated between the optical module and the peer end is used for precise error correction, thereby avoiding the computational crash caused by directly sending erroneous data packets to the supercomputing host.
[0083] The closed-loop feedback control is triggered, and the abnormal state corresponding to the abnormal data is fed back to the digital twin link. The modulation order, transmit power and routing mapping relationship in the optical layer communication strategy are adjusted in reverse. Among them, the internal bus of the optical module packages the current "abnormal state, including degradation mark, degradation occurrence timestamp and specific over-limit parameter value" into a status feedback message, and transmits it back to the digital twin space constructed in the aforementioned step S121 along the high-speed control interface. After receiving the feedback, the digital twin link immediately triggers the emergency replay mechanism.
[0084] Based on the abnormal OSNR and EVM values reported, the internal channel state baseline is updated, and the heuristic multi-objective joint optimization algorithm is restarted. In this reverse simulation, the action space is forcibly changed: in terms of modulation order, it is downgraded from the original 16QAM to QPSK to expand the constellation point distance and improve the anti-crosstalk tolerance; in terms of transmit power, the gain of the digital-to-analog converter (DAC) or the bias current of the adjustable optical amplifier is dynamically adjusted, for example, increasing the transmit optical power from 2.5dBm to 3.5dBm to offset the signal-to-noise ratio degradation, or reducing the power when the nonlinear constraints are not met; in terms of routing mapping, the mapping dictionary of the spatial multiplexing channel is adjusted, and the traffic flow of "Fiber Core 1" which is severely affected by transient crosstalk is rerouted to the backup, well-maintained "Fiber Core 5". The digital twin link outputs a brand-new optical layer communication strategy.
[0085] Please see Figure 2 The specific application scenarios and corresponding embodiments of this application are as follows: In step S111, it is assumed that in the network communication scenario of the supercomputing center, optical modules are deployed as interconnection hardware between nodes. When node A initiates a gradient data synchronization communication request for large-scale distributed intelligent training to node B: the data receiver corresponding to the optical module captures the Ethernet physical layer data stream carrying the request; the data receiver performs deep protocol parsing on the received data frame, sequentially stripping the Ethernet frontend header and IP header. When parsing to the transport layer and application layer, it identifies the UDP destination port number as 4791, and parses the opcode "AtomicWrite" or "Send" in the Infiniband header, thus determining that the network communication request belongs to RDMA strong consistency write operation traffic.
[0086] Based on the above analysis results, the data receiver extracts communication requirement indicators: Since it is confirmed to be an RDMA core operation, the local SLA mapping table is queried, and the extracted end-to-end latency threshold is determined to be 0.8μs; since the RDMA protocol layer does not contain a complex retransmission tolerance mechanism, the extracted packet loss rate sensitivity level is directly marked as "extremely high sensitivity".
[0087] Meanwhile, by statistically analyzing the physical layer coding rate and packet interval of the packets in the stream, it was found that the burst transmission characteristics occurred with the alternation of GPU computing cycles, and the bandwidth fluctuation range was calculated to be 50Gbps to 200Gbps. Through this series of staged parsing and extraction, the optical module directly obtained the full-domain service intent characteristics of the intelligent training traffic at the underlying hardware level, providing a precise service constraint benchmark for the synchronous collection and joint optimization of the underlying physical multi-dimensional operating parameters in the subsequent step S112.
[0088] In step S112, when the optical module is executing the large-scale distributed intelligent training gradient data synchronization communication request determined in step S111, the micro-sensor array and high-speed sampling circuit inside the optical module are synchronously activated.
[0089] During the acquisition of the physical status of the link, the server cluster fans in the supercomputing center's computer room caused high-frequency micro-vibration of the rack, resulting in a physical micro-bend across the multi-core fiber. At this time, the optical module's transmitter injected a low-frequency pilot signal into one fiber core. Due to the micro-bend of the fiber causing mode coupling, the pilot signal leaked to the adjacent fiber core. The pilot signal detection unit integrated at the edge of the adjacent fiber core captured this abnormally enhanced pilot leakage signal. After DSP calculation inside the module, it was found that the dynamic channel crosstalk value between the adjacent fiber cores across the multi-core fiber deteriorated from the usual -35dB to -28dB.
[0090] Meanwhile, due to the deterioration of this crosstalk value and the additional loss caused by tiny dust particles in the optical connector, the signal power statistics collected by the optical receiver fluctuated; the DSP inside the optical module analyzed the bit error distribution characteristics before FEC decoding and found that the bit error rate dropped from 10-15 Rise to 10 -9 Furthermore, the variance increases; combining the signal power and the error distribution characteristics, the SNR-BER mapping model is used to calculate that the current optical transceiver signal-to-noise ratio has decreased from 28dB to 24dB.
[0091] In addition, since the ambient temperature in the supercomputing room has risen by 10 degrees Celsius since the cold start, the thermo-optical effect of the silicon photonic modulator inside the optical module has caused its refractive index to drift. After the high-speed sampling circuit detects the slight jitter of the DC bias voltage, it plots the dynamic response curve of the optical output power and compares it with the ideal sine curve. It is determined that the bias point of the silicon photonic modulator has drifted to the right by 0.15 volts, that is, the bias point drift is 0.15V.
[0092] The optical module fuses and encapsulates the end-to-end latency threshold (0.8μs), extremely high packet loss rate sensitivity level, and bandwidth fluctuation range (50Gbps to 200Gbps) of the acquired intelligent training traffic with the dynamic channel crosstalk value (-28dB), optical transceiver signal-to-noise ratio (24dB), and offset point drift (0.15V) at this moment to construct a full-domain perception dataset for the current moment. This dataset is then sent to the subsequent digital twin link model for policy pre-playing. Through this series of staged physical perception and data fusion, the optical module achieves synchronous quantification of physical communication channel degradation and service intent.
[0093] In step S121, the specific steps are as follows: Assuming that in the network communication scenario of the supercomputing center, an optical module is deployed as hardware for interconnection between nodes based on cross-core optical fiber; when the optical module is executing AI gradient data synchronization service, the microcontroller inside the optical module starts the digital twin construction process in step S121.
[0094] The optical module takes the "intelligent training traffic characteristics: end-to-end latency threshold 0.8μs, extremely high packet loss rate sensitivity level, bandwidth fluctuation range 50G to 200G" parsed in step S111 as a logical message and sends it to the top constraint layer of the digital twin space embedded in the module along the control bus; at the same time, it packs the "physical layer degradation status: dynamic channel crosstalk value -28dB, optical transceiver signal-to-noise ratio 24dB, offset point drift amount 0.15V" collected in step S112 into a high-speed sampling data stream and sends it to the bottom physical modeling layer of the digital twin space along the data bus.
[0095] The digital twin space undergoes multi-level fusion internally. At the feature level, the twin space aligns the -28dB crosstalk value with the 24dB signal-to-noise ratio and generates a set of complex feature vectors by combining the 0.15V bias drift. At the model level, the twin modeling engine extracts this feature vector, calls the internal nonlinear Schrödinger equation solver, substitutes -28dB into the mode coupling coefficient matrix in the multi-core fiber boundary condition, and substitutes 0.15V into the silicon photonic modulator transfer function model to update the phase shift of the modulation arm.
[0096] In this process, the "0.8μs delay threshold" issued by the top constraint layer is transformed into a limit on the step size of dispersion compensation calculation in the model, and the "extremely high packet loss rate sensitivity level" is transformed into a strict constraint on the upper limit of nonlinear noise power accumulation; the two types of models are deeply coupled in the iterative process of solving the equations, generating the digital twin link of the current optical module.
[0097] Through the above fusion, the twin link can map in real time the dynamic evolution of crosstalk between adjacent fiber cores caused by micro-bending of the multi-core fiber and the real environmental state of the silicon optical waveguide undergoing a slight shift due to the nonlinear Kerr effect caused by the temperature rise when the fan vibrates and the temperature rises. This twin link, which evolves synchronously with the physical link, allows the subsequent step S122 (i.e., joint optimization strategy pre-simulation) to be tested and solved in a virtual sandbox that perfectly matches the harsh physical environment of the current supercomputing center.
[0098] In step S122, an optical module is deployed as hardware for inter-node interconnection based on multi-core optical fiber. When the optical module is performing gradient data synchronization service, its internal digital twin link has mapped the current channel crosstalk value deteriorated to -28dB, the optical signal-to-noise ratio dropped to 24dB, and the silicon photonic modulator bias drifted by 0.15V, and has obtained the service requirements that the end-to-end delay threshold must be less than 0.8μs and the sensitivity level of extremely high packet loss rate.
[0099] At this point, the joint optimization mechanism in step S122 is triggered; the cost function transforms the extremely high packet loss rate sensitivity level into a constraint, requiring that the predicted Pre-FEC bit error rate of the twin link must be lower than 10 during the simulation process, regardless of the parameter combination used. -3 To ensure zero packet loss; at the same time, to maximize bandwidth utilization and minimize latency to less than 0.8μs.
[0100] The digital twin space has begun multiple rounds of strategy rehearsals; assuming the first round of traversal testing used a combination of "64QAM modulation + FEC coding rate 0.9 + transmit optical power 3dBm", the twin link, based on a crosstalk environment of -28dB, extrapolated a bit error rate as high as 10%. -4 This combination violated the link stability constraint and was therefore removed.
[0101] The second round of traversal testing tested the combination of "QPSK modulation + FEC coding rate 0.8". Although the bit error rate met the requirements, the low-order modulation resulted in excessively low spectral efficiency, leading to a poor bandwidth utilization objective function. Furthermore, the low FEC coding rate increased processing latency, causing the end-to-end latency to exceed 0.8 μs. After multiple rounds of heuristic search, the algorithm gradually converged to a combination: "16QAM + FEC coding rate 0.85 + optical power 2.5 dBm" was used for "fiber core 1" with less crosstalk, while "QPSK + FEC coding rate 0.9" was downgraded for "fiber core 2" with micro-bend crosstalk. This combination yielded a predicted bit error rate of 5 × 10⁻⁶ in the twin link. -5 It satisfies the stability constraints, and the overall bandwidth utilization and latency reach Pareto optimality, thus satisfying the convergence condition.
[0102] The optical module encapsulates the optimal solution into an optical layer communication strategy and outputs it. This optical layer communication strategy includes an adaptive resource configuration table, indicating that fiber core 1 is allocated the main data stream and uses 16QAM, fiber core 2 is allocated the auxiliary verification stream and uses QPSK, and a set of signal processing parameters, including: the target adjustment value of the bias voltage that needs to compensate for 0.15V drift, the DAC sampling rate setting corresponding to 16QAM, and the tap coefficient of the receiver DSP nonlinear equalization compensation designed for -28dB crosstalk. This strategy will be directly used in the subsequent step S13 to drive the silicon photonics integrated engine for hardware reconstruction.
[0103] In step S131, the specific steps are as follows: Assuming that in the network communication scenario of the supercomputing center, optical modules are deployed as hardware for interconnection between nodes based on multi-core optical fibers; when the optical modules are performing AI gradient data synchronization services, the aforementioned step S122 has generated the optimal optical layer communication strategy, which stipulates that: for "fiber core 1" with less crosstalk, a configuration of "16QAM modulation + FEC coding rate 0.85" is adopted, and the signal processing parameter set requires compensation for the 0.15V bias point drift.
[0104] Based on this, the microcontroller inside the optical module encapsulates this strategy into a control command stream. For example, it generates a control sequence containing a hexadecimal instruction packet header, where specific fields configure the modulation format register to "0011" to represent 16QAM, the FEC coding rate register to the denominator and numerator parameters corresponding to "0.85", and the DAC gain control word to the voltage amplitude corresponding to the optimal eye diagram opening degree of 16QAM. This control command stream is burst-down to the internal control register array of the silicon photonics integrated engine of the optical module through the PCIe or high-speed SPI interface inside the optical module in milliseconds.
[0105] The silicon photonics integrated engine then analyzes the instruction stream; for signal amplitude adjustment, the internal digital-to-analog converter (DAC) reads the gain control word and dynamically adjusts the output differential voltage swing from 1.5V in the original QPSK mode to 0.8V adapted to 16QAM precision modulation, in order to avoid nonlinear clipping distortion caused by overload of the drive signal under high-order modulation; for phase adjustment, the engine reads the bias drift compensation parameter and dynamically corrects the DC bias voltage applied to the Mach-Zehnder modulator interferometer arm from the standard -1.0V to -1.15V through the internally integrated DAC channel, in order to compensate for the 0.15V offset caused by temperature rise, so that the modulator's operating point is precisely locked back to the quadrature bias point.
[0106] After completing the above simulation parameter configuration, the digital logic unit inside the engine starts; the received intelligent training gradient data stream enters the FEC encoder, and at a coding rate of 0.85, approximately 17.6 parity bits are inserted after every 100 information bits; the parity bit stream enters the constellation mapping unit, and according to the 16QAM format, every 4 bits are grouped and mapped to 16 specific amplitude and phase coordinate points on the complex plane. This I / Q coordinate data is converted into differential analog electrical signals by the configured digital-to-analog converter and driving circuit, and finally loaded onto the RF electrodes of the silicon photonic modulator to jointly modulate the amplitude and phase of the continuous optical carrier from the laser; at this point, the optical module has successfully converted the gradient data stream in the electrical domain into an optical signal carrying data that conforms to the 16QAM modulation characteristics and has error correction redundancy with a coding rate of 0.85, completing the electro-optic signal conversion, and is ready to inject it into the spatially multiplexed target optical communication path.
[0107] In step S132, the specific steps are as follows: Assuming that in the network communication scenario of the supercomputing center, optical modules are deployed as hardware for interconnection between nodes based on seven-core and multi-core optical fibers; when the optical modules are performing large-scale AI gradient data synchronization services, the optical layer communication strategy output in the aforementioned step S122 indicates that: in order to avoid physical micro-bending crosstalk caused by micro-vibration of the computer room fan, the “fiber core 4” located in the center should be avoided, as its crosstalk value has deteriorated to -28dB, and the service flow should be split, with the main data flow mapped to the edge “fiber core 1” and using 16QAM modulation, and the verification data flow mapped to “fiber core 3” and using QPSK modulation; at the same time, both data flows need to occupy wavelength division multiplexing channels with wavelengths of “λ1: 1550.0nm” and “λ2: 1550.9nm”, respectively.
[0108] Based on this, the optical module obtains the mapping relationship of the spatial multiplexing channel; the microcontroller reads the correspondence between the logical flow and the physical fiber core from the resource configuration table, confirms that the physical route of fiber core 1 and fiber core 3 needs to be activated, and blocks the path to fiber core 4.
[0109] The silicon photonics integrated engine controls the internal silicon-based 2×2 optical switch matrix based on this mapping relationship. The engine applies a bias voltage to specific switch nodes, changing the refractive index of their internal waveguides. This guides the main data optical signal, which has already undergone 16QAM electro-optic conversion, to the output waveguide of the corresponding fiber core 1, and guides the verification data optical signal, which has undergone QPSK electro-optic conversion, to the output waveguide of the corresponding fiber core 3. These two optical signals enter the multi-core fiber fan-in / fan-out device of the optical module. The fan-in / fan-out device uses its internal three-dimensional curved microlens array to focus and calibrate the angles of the two independent optical signals, ensuring that they are precisely aligned with the incident cross-sections of fiber cores 1 and 3 on the splice end face of the multi-core fiber in physical space. This achieves physical coupling of optical signals in the spatial division multiplexing dimension with extremely low insertion loss.
[0110] During this process, the multidimensional multiplexing architecture operates synchronously; the arrayed waveguide grating (AWG) inside the optical module has already combined the two wavelengths λ1 and λ2 in the frequency domain and combined them with the orthogonal polarization state generated by the polarization beam splitter (PBS) and injected them into the output waveguide; therefore, when the optical signal enters fiber core 1 and fiber core 3 through the fan-in and fan-out devices, it actually carries a multidimensional signal of wavelength division multiplexing and polarization multiplexing inside each fiber core; thus, the optical module has successfully established a target optical communication path that avoids the damaged fiber core 4, is jointly carried by fiber core 1 and fiber core 3, and integrates wavelength and polarization multidimensional multiplexing, transmitting the AI gradient data stream to the supercomputing node peer with extremely high physical parallelism.
[0111] In step S141, an optical module is deployed as hardware for inter-node interconnection based on multi-core optical fiber. When the optical module is transmitting a large-scale AI gradient synchronization data stream via the target optical communication path "fiber core 1" established in step S132, a sudden high-frequency mechanical vibration of a computing node in the supercomputing room causes a surge in transient mode coupling crosstalk between "fiber core 1" and adjacent fiber cores. At this time, the monitoring mechanism of step S141 is triggered.
[0112] The performance monitoring module integrated inside the silicon photonics receiving engine of the optical module extracts 5% of the optical power from the 16QAM master data optical signal being received as an in-band monitoring signal through an optical coupler and imports it into the DSP unit. The spectrum analysis module of the DSP unit performs FFT on the signal and finds that the signal optical power at the center wavelength remains unchanged, but the out-of-band noise floor increases by 2dB due to the surge in crosstalk. The calculation shows that the real-time estimate of the current optical signal-to-noise ratio further decreases from the aforementioned 24dB to 22dB.
[0113] Meanwhile, after the constellation recovery algorithm inside the DSP completes synchronization, it extracts the actual coordinates of the received 16QAM symbol and compares them with the ideal 16QAM reference point. Due to the combined effect of transient crosstalk and link residual dispersion, the actual symbol point diverges severely in the complex plane, and the calculated error vector magnitude (EVM) deteriorates from the normal 8% to 15%.
[0114] Furthermore, before sending the data to the soft-decision FEC decoder, the DSP uses a built-in statistical counter to compare the hard-decision bits of the previous 10,000 frames of signal with the local checksum; the statistical results show that the current Pre-FEC bit error rate has been reduced from 5×10⁻⁶. -5 Jump to 2×10 -3 These three real-time transmission quality parameters are: OSNR 22dB, EVM 15%, and Pre-FECBER 2×10. -3 "The data is synchronously packaged and marked with a timestamp, which directly serves as the core criterion for subsequent step S142 to determine whether the reverse adjustment mechanism is triggered. Through this in-band real-time high-frequency detection, the optical module can capture the severe degradation trend of the physical link several milliseconds in advance, before the FEC error correction margin is exhausted and actual service packet loss occurs."
[0115] In step S142, optical modules are deployed as hardware for inter-node interconnection across multi-core optical fibers. While the optical modules are performing large-scale gradient synchronization and transmitting 16QAM optical signals via "fiber core 1," step S141 detects a surge in transient crosstalk in "fiber core 1" due to high-frequency mechanical vibration in the equipment room. This causes the real-time OSNR to drop to 22dB, the EVM to deteriorate to 15%, and the Pre-FEC bit error rate to jump to 2×10⁻⁶. -3 .
[0116] At this point, the current Pre-FEC bit error rate (2×10) -3 The error exceeded the preset quality parameter threshold (1×10) set for the sensitivity level of extremely high packet loss rate in intelligent training. -3 The microcontroller immediately outputs a hardware interrupt and initiates an exception handling mechanism. At the control channel level, the optical module inserts a specific degradation marker into the out-of-band Ethernet overhead frame accompanying the gradient data stream. For example, setting a certain control field to "0xFF" indicates physical layer degradation. At the same time, the FIFO buffer inside the optical module is activated, which intercepts and buffers the data frames corresponding to the 16QAM symbols that have diverged due to crosstalk during this period. The data frames are not sent to the GPU memory of the supercomputing node for the time being, but wait for the link to self-heal before the underlying layer corrects or retransmits them.
[0117] At the same time, closed-loop feedback control is triggered; the microcontroller sends an abnormal status message containing information such as "OSNR 22dB" and "Fiber core 1 is subject to strong crosstalk" back to the digital twin link of the optical module via the internal bus; after receiving the abnormal status, the twin link immediately pauses the current strategy and performs joint optimization re-run in the virtual sandbox; the re-run algorithm determines that 16QAM cannot converge at the current signal-to-noise ratio of 22dB, and thus reverses the modulation order from 16QAM to the more robust QPSK.
[0118] Meanwhile, to compensate for the decrease in signal-to-noise ratio, the transmit power was increased from 2.5dBm to 3.2dBm. More importantly, regarding the routing mapping relationship, the twin link determined that "fiber core 1" was unavailable and switched the spatial multiplexing route from "fiber core 1" to "fiber core 5", which is far away from the vibration source and is currently in good condition. This series of reverse adjustments generated a brand-new optical layer communication strategy, which will be quickly sent to the silicon photonics integrated engine of the optical module to drive the physical layer to reconfigure the electro-optical parameters and optical switch matrix, so that the intelligent training data stream between supercomputing nodes can adaptively cross physical fault nodes without interrupting the service.
[0119] Please see Figure 3 The optical module-based network communication system is applied to the aforementioned optical module-based network communication method; the optical module-based network communication system includes: The parameter module 21 is used to collect the corresponding network communication requests, determine the corresponding communication requirement indicators based on the parsing of the network communication requests, and simultaneously collect the multi-dimensional operating status parameters of the optical module under the current link. The multi-dimensional operating status parameters include the optical transceiver signal-to-noise ratio, offset point drift, and dynamic channel crosstalk value across multi-core optical fibers of the optical module. Strategy module 22 is used to construct a digital twin link of the optical module based on communication demand indicators and multi-dimensional operating status parameters. In the digital twin link, the joint optimization objective is to maximize bandwidth utilization and minimize end-to-end latency. The strategy pre-plays the joint optimization objective to generate the corresponding optical layer communication strategy. The communication path module 23 is used to send the optical layer communication strategy to the silicon photonics integrated engine of the optical module, drive the silicon photonics integrated engine to complete the electro-optic signal conversion according to the target modulation format and forward error correction code coding rate, and establish the target optical communication path according to the mapping relationship of the spatial multiplexing channel. The adjustment module 24 is used to monitor the transmission quality parameters of the network communication in real time during the transmission of data stream through the target optical communication path. When the transmission quality parameters are lower than the preset quality parameter threshold, the corresponding abnormal data is marked and the optical layer communication strategy of the optical module is adjusted in reverse.
[0120] It should be noted that although multiple modules are mentioned in the detailed description above, this division is not mandatory; the above description is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the protection scope of the present invention; therefore, the protection scope of the present invention should be determined by the protection scope of the claims.
Claims
1. A network communication method based on optical modules, characterized in that, include: Collect the corresponding network communication requests, determine the corresponding communication requirement indicators based on the parsing of the network communication requests, and simultaneously collect the multi-dimensional operating status parameters of the optical module under the current link. The multi-dimensional operating status parameters include the optical transceiver signal-to-noise ratio, offset point drift, and dynamic channel crosstalk value across multi-core optical fibers of the optical module. A digital twin link for the optical module is constructed based on communication demand indicators and multi-dimensional operating status parameters. In the digital twin link, the joint optimization objective is to maximize bandwidth utilization and minimize end-to-end latency. The strategy is pre-drilled for the joint optimization objective to generate the corresponding optical layer communication strategy. The optical layer communication strategy is sent to the silicon photonics integrated engine of the optical module, which drives the silicon photonics integrated engine to complete the electro-optical signal conversion according to the target modulation format and forward error correction code coding rate, and establishes the target optical communication path according to the mapping relationship of the spatial multiplexing channel. During the transmission of data streams through the target optical communication path, the transmission quality parameters of network communication are monitored in real time. When the transmission quality parameters are lower than the preset quality parameter threshold, the corresponding abnormal data is marked, and the optical layer communication strategy of the optical module is adjusted in reverse.
2. The network communication method based on optical modules according to claim 1, characterized in that, The process involves collecting corresponding network communication requests, determining corresponding communication requirement indicators based on the parsing of these requests, and simultaneously collecting multi-dimensional operational status parameters of the optical module under the current link. These multi-dimensional operational status parameters include the optical transceiver signal-to-noise ratio, offset point drift, and dynamic channel crosstalk value across multi-core optical fibers, including: Based on the network communication requests collected by the data receiving end, the network communication requests are subjected to in-depth protocol parsing, and communication requirement indicators at the business level are extracted. These indicators include preset end-to-end latency thresholds, packet loss rate sensitivity levels, and bandwidth fluctuation ranges.
3. The network communication method based on optical modules according to claim 2, characterized in that, The process involves collecting corresponding network communication requests, determining corresponding communication requirement indicators based on the parsing of these requests, and simultaneously collecting multi-dimensional operational status parameters of the optical module under the current link. These multi-dimensional operational status parameters include the optical transceiver signal-to-noise ratio, offset point drift, and dynamic channel crosstalk value across multi-core optical fibers of the optical module. It also includes: By deploying a miniature sensor array and high-speed sampling circuit inside the optical module, multi-dimensional operating status parameters of the optical module under the current link are collected synchronously. Among them, the pilot signal detection unit integrated on the edge of the multi-core fiber is used to measure and calculate the dynamic channel crosstalk value between adjacent cores in the multi-core fiber in real time. Combined with the signal power statistics and bit error distribution characteristics of the optical transceiver, the optical signal-to-noise ratio is calculated. By monitoring the dynamic response curve of the DC bias voltage and optical output power of the silicon photonic modulator, the bias point drift is calibrated, and a global perception dataset including service intent and physical layer status is constructed.
4. The network communication method based on optical modules according to claim 1, characterized in that, The digital twin link of the optical module is constructed based on communication demand indicators and multi-dimensional operating status parameters. In the digital twin link, maximizing bandwidth utilization and minimizing end-to-end latency are the joint optimization objectives. A strategy pre-playing is performed on the joint optimization objectives to generate the corresponding optical layer communication strategy, including: Communication demand indicators and multi-dimensional operating status parameters are input into the same digital twin space along different paths, and multi-level fusion is carried out in the digital twin space to gradually build the digital twin link of the optical module. The digital twin link can map the nonlinear effects and multi-core crosstalk environment of the current optical channel in real time.
5. The network communication method based on optical modules according to claim 4, characterized in that, The digital twin link of the optical module is constructed based on communication demand indicators and multi-dimensional operating status parameters. In the digital twin link, the joint optimization objective is to maximize bandwidth utilization and minimize end-to-end latency. The strategy is pre-drilled on the joint optimization objective to generate the corresponding optical layer communication strategy. The method also includes: A joint optimization mechanism is introduced into the digital twin link, with the joint optimization objectives of maximizing bandwidth utilization and minimizing end-to-end latency, and link stability as a constraint. The joint optimization objectives are subjected to multiple rounds of strategy pre-playing, traversing different modulation formats, forward error correction coding rates and power configuration combinations in the digital twin space, so as to output the optimal solution that meets the preset convergence conditions, and use it as the optical layer communication strategy. The optical layer communication strategy includes an adaptive resource configuration table for different multi-core fiber cores and the corresponding signal processing parameter set.
6. The network communication method based on optical modules according to claim 1, characterized in that, The process of disseminating the optical layer communication strategy to the silicon photonics integrated engine of the optical module, driving the silicon photonics integrated engine to complete the electro-optical signal conversion according to the target modulation format and forward error correction code coding rate, and establishing the target optical communication path based on the mapping relationship of the spatial multiplexing channel, includes: The optical layer communication strategy is encapsulated into a corresponding control command stream and sent to the silicon photonics integrated engine of the optical module through a high-speed control interface. The silicon photonics integrated engine parses the control command stream, dynamically configures its internal digital-to-analog converter and drive circuit, thereby adjusting the amplitude and phase of the drive signal, loading the constellation map mapping according to the target modulation format, and completing the corresponding electro-optic signal conversion in conjunction with the target forward error correction code coding rate.
7. The network communication method based on optical modules according to claim 6, characterized in that, The process of distributing the optical layer communication strategy to the silicon photonics integrated engine of the optical module, driving the silicon photonics integrated engine to complete the electro-optical signal conversion according to the target modulation format and forward error correction code coding rate, and establishing the target optical communication path based on the mapping relationship of the spatial multiplexing channel, further includes: The mapping relationship of the spatial division multiplexing channel is obtained, and the optical switch matrix and multi-core fiber fan-in and fan-out devices are controlled based on the mapping relationship to couple the loaded optical signal to a specific core of the target multi-core fiber. The corresponding target optical communication path is established by combining the wavelength division multiplexing mechanism and the multi-dimensional multiplexing architecture.
8. The network communication method based on optical modules according to claim 1, characterized in that, During the transmission of data streams through the target optical communication path, the transmission quality parameters of network communication are monitored in real time. When the transmission quality parameters are lower than a preset quality parameter threshold, the corresponding abnormal data is marked, and the optical layer communication strategy of the optical module is adjusted in reverse, including: During the transmission of data streams through the target optical communication path, the performance monitoring module integrated within the optical module performs real-time in-band detection of the signal to obtain the transmission quality parameters of the network communication. These transmission quality parameters include the real-time estimate of the optical signal-to-noise ratio, the magnitude of the error vector, and the bit error rate statistics before soft-decision forward error correction decoding.
9. The network communication method based on optical modules according to claim 8, characterized in that, The method of monitoring network communication transmission quality parameters in real time during data flow transmission through the target optical communication path, marking corresponding abnormal data when the transmission quality parameters are lower than a preset quality parameter threshold, and adjusting the optical layer communication strategy of the optical module in reverse, further includes: The transmission quality parameters are compared with preset quality parameter thresholds in real time. When the transmission quality parameters are lower than the preset quality parameter thresholds, an anomaly handling mechanism is activated. Specific degradation markers are inserted into the control channel to mark the corresponding abnormal data, and the abnormal data is cached for subsequent retransmission or error correction. At the same time, closed-loop feedback control is triggered to feed back the abnormal status corresponding to the abnormal data to the digital twin link, and the modulation order, transmit power and routing mapping relationship in the optical layer communication strategy are adjusted in reverse.
10. A network communication system based on optical modules, characterized in that, The optical module-based network communication system is applied to the optical module-based network communication method as described in any one of claims 1-9; the optical module-based network communication system includes: The parameter module is used to collect the corresponding network communication requests, determine the corresponding communication requirement indicators based on the parsing of the network communication requests, and simultaneously collect the multi-dimensional operating status parameters of the optical module under the current link. The multi-dimensional operating status parameters include the optical transceiver signal-to-noise ratio, offset point drift, and dynamic channel crosstalk value across multi-core optical fibers of the optical module. The strategy module is used to construct a digital twin link of the optical module based on communication demand indicators and multi-dimensional operating status parameters. In the digital twin link, the joint optimization objective is to maximize bandwidth utilization and minimize end-to-end latency. The strategy pre-plays the joint optimization objective to generate the corresponding optical layer communication strategy. The communication path module is used to send the optical layer communication strategy to the silicon photonics integrated engine of the optical module, drive the silicon photonics integrated engine to complete the electro-optic signal conversion according to the target modulation format and forward error correction code coding rate, and establish the target optical communication path according to the mapping relationship of the spatial multiplexing channel. The adjustment module is used to monitor the transmission quality parameters of network communication in real time during the transmission of data streams through the target optical communication path. When the transmission quality parameters are lower than the preset quality parameter threshold, the corresponding abnormal data is marked, and the optical layer communication strategy of the optical module is adjusted in reverse.
