Self-learning MCF-SDM long-distance optical communication system nonlinear compensation method and device

Abstract

The invention provides a self-learning MCF-SDM long-distance system non-linear compensation method, which improves the performance of system non-linear compensation, pre-executes linear compensation, stores the filter coefficient of each linear damage compensation module obtained by calculation during pre-training, divides the one-dimensional signal sequence of each channel of original data into a plurality of blocks, and performs non-linear compensation. Adding adjacent data to the head and tail positions of each block and mapping the data to a two-dimensional structure, and adding channel dimensions to form three-dimensional input data; constructing an LDBP compensation module comprising a plurality of dispersion compensation layers and nonlinear compensation layers which are alternately arranged, and sending the three-dimensional input data into the LDBP compensation module and other linear damage compensation modules for training until a trained LDBP compensation module is obtained; and sequentially inputting to-be-processed data into other linear damage compensation modules except the trained LDBP compensation module and the dispersion compensation module, and performing damage compensation on the data.

Description

Technical Field

[0001] The present invention relates to the technical field of optical fiber communication, and in particular to a self-learning nonlinear compensation method and device for a long-distance MCF-SDM optical communication system. Background Art

[0002] The rapid growth of global data traffic has driven the increasing demand for higher bandwidth in optical communication systems. In the past 20 years, the proposed polarization division multiplexing (PDM), wavelength division multiplexing (WDM), and high-order modulation technologies have utilized all degrees of freedom of single-mode fibers, and the transmission capacity of traditional single-mode fibers has approached the Shannon limit. Space division multiplexing (SDM) technology based on multi-core fiber (MCF) or few-mode fiber (FMF) can simultaneously transmit multiple spatial channels in a single fiber, doubling the system capacity, and has become a promising solution to this challenge. However, fiber nonlinearity remains the fundamental limit to the improvement of the capacity of contemporary long-distance optical transmission systems. The interaction between dispersion (CD), crosstalk (XT), polarization mode dispersion (PMD) and nonlinear effects among different spatial channels in SDM fibers makes the nonlinear effects more complex, posing a greater challenge to nonlinear compensation.

[0003] Researchers in the prior art have proposed various algorithms to mitigate fiber nonlinear distortion. The digital backpropagation (DBP) algorithm alternately compensates for dispersion (CD) and nonlinear impairments based on the split-step Fourier method (SSFM). However, the actual implementation of DBP requires a complete understanding of the fiber link parameters, which is infeasible in real-world scenarios. In recent years, the rapid development of machine learning (ML) has received extensive attention due to its excellent ability to solve complex problems. Artificial neural networks (ANN) and long short-term memory (LSTM) networks have been introduced into the field of fiber nonlinear compensation. However, these machine learning-based methods are "black box" models, lacking interpretability and having a high computational complexity, making it difficult to apply in engineering. In the MCF-SDM (multi-core fiber space division multiplexing) long-distance optical fiber communication system, the traditional method does not consider the impact of crosstalk between different spatial channels on the nonlinear effect, and the method is not applicable to the MCF-SDM (multi-core fiber space division multiplexing) system. Summary of the Invention

[0004] The object of the present invention is to overcome the deficiencies of the prior art and provide a self-learning nonlinear compensation method for a long-distance MCF-SDM system. Aiming at the nonlinear impairments in the MCF-SDM long-distance optical communication system, by constructing coefficients characterizing the influence of crosstalk and polarization mode dispersion on the nonlinear noise during the LDBP process to calculate the nonlinear phase, the performance of nonlinear compensation for the MCF-SDM long-distance optical communication system is improved.

[0005] The technical solution is as follows: A self-learning nonlinear compensation method for MCF-SDM long-distance systems, characterized by including the following steps:

[0006] Step 1: Use known transmitted data and received data, pre-perform linear compensation through a linear impairment compensation module, and save the filter coefficients of each linear impairment compensation module calculated during pre-training. The linear impairment compensation module includes a dispersion compensation module, a phase offset compensation module, a frequency offset compensation module, and a multi-input multi-output linear equalization module;

[0007] Step 2: Divide the one-dimensional signal sequence of each channel of the original data into multiple blocks, add adjacent data to the head and tail positions of each block and map it into a two-dimensional structure, add the channel dimension to form three-dimensional input data;

[0008] Step 3: Construct an LDBP compensation module for dispersion compensation. The LDBP compensation module includes several alternately arranged dispersion compensation layers and nonlinear compensation layers. The dispersion compensation layer performs dispersion compensation using a complex FIR filter in the time domain, and processes the signals of each fiber core in the time domain respectively; the nonlinear compensation layer calculates the nonlinear phase to be compensated for each polarization channel of each fiber core using the amplitudes of the signals from all fiber cores, and performs nonlinear phase compensation in the time domain through an exponential function;

[0009] Step 4: Use the LDBP compensation module to replace the dispersion compensation module, send the three-dimensional input data formed in Step 2 into other linear impairment compensation modules except the LDBP compensation module and the dispersion compensation module for training, freeze the filter coefficients in other linear impairment compensation modules, and use the gradient descent method to optimize the coefficients of the LDBP compensation module until a trained LDBP compensation module is obtained;

[0010] Step 5: Sequentially input the data to be processed into other linear impairment compensation modules except the trained LDBP compensation module and the dispersion compensation module to perform impairment compensation on the data. Among them, freeze the filter coefficients of the trained LDBP compensation module, and adaptively update the filter coefficients of other linear impairment compensation modules.

[0011] Further, in Step 2, when adding adjacent data to the head and tail positions of each block, for the first block and the last block, add additional zero elements at the starting position of the first block and the ending position of the last block respectively to align the data.

[0012] Further, the two-dimensional structure of the data in Step 2 is represented by the following matrix:

[0013]

[0014] Among them, one row in the two-dimensional matrix corresponds to one block, the number of columns in the two-dimensional matrix represents the amount of data within the block, b is the size of each block, and nb is the total length of the one-dimensional signal sequence data; s is the length of the added extra adjacent data; k is the channel number, representing the fiber core or polarization; adding the channel dimension forms three-dimensional input data, and the channel dimension includes the fiber core and polarization. The polarization and the fiber core are treated as equivalent channels and processed in parallel.

[0015] Further, in step 3, the amplitude of the signal of polarization x or polarization y in fiber core k after propagating a distance h in the time domain is expressed in the following form:

[0016] u k,x / y (h,t) = F -1 [H(h) × F(u k,x / y (0,t))]

[0017] Among them, H(h) represents the transfer function in the frequency domain, F -1 represents the Fourier transform, t represents time, and F(u k,x / y (0,t)) represents the frequency-domain signal obtained by performing a Fourier transform on the initial time-domain signal when the propagation distance is 0 and the polarization is x or y in fiber core k; dispersion compensation is effectively achieved through the frequency-domain compensation operator H -1 (h). The dispersion compensation layer uses a complex FIR filter in the time domain to approximate the frequency-domain compensation operator H -1 (h) for dispersion compensation, and processes the signals of each fiber core in the time domain separately. The output of the dispersion compensation layer is shown in the following formula:

[0018]

[0019] Among them, x x∣yk (i) represents the uncompensated original signal of polarization x or polarization y in fiber core k, z x∣yk (i) represents the output of the dispersion compensation layer of polarization x or polarization y in fiber core k, and i is the signal sequence number; w k is the weight of the FIR filter, and S is the width of the FIR filter.

[0020] Further, in step 3, the nonlinear compensation layer calculates the nonlinear phase that needs to be compensated for each polarization channel of each fiber core using the amplitudes of the signals from all fiber cores, and performs nonlinear phase compensation in the time domain through an exponential function.

[0021] Further, in step 3, the nonlinear compensation layer calculates the nonlinear phase that needs to be compensated for each polarization channel of each core using the amplitudes of the signals from all fiber cores, and directly performs nonlinear phase compensation in the time domain through an exponential function. The calculated nonlinear phase is shown in the following formula:

[0022] f(z x∣yk (i)) = exp[-iΦ x∣yk (i)]z x∣yk (i)

[0023]

[0024] where z x∣yk (i) represents the output of the dispersion compensation layer with polarization x or polarization y in core k, α xkp and α ykp are trainable parameters, i is the signal number, C is the total number of cores; p is the core number, and core p is the number of another core interacting with core k, with the range [1, C].

[0025] Furthermore, in step 4, the gradient descent method is used to optimize the coefficients of the LDBP compensation module. Through the error between the center b data of each channel in the output of the last linear compensation filter in other linear damage compensation modules and the known transmitted signal for training, all coefficients in the LDBP compensation module are updated through gradient backpropagation. The loss function is:

[0026]

[0027] where L MSE is the minimum mean square error value, H xij is the signal output from the last linear compensation filter after the output of the LDBP compensation module is input into other linear compensation modules, T xij is the known transmitted data, and C is the total number of cores.

[0028] Compared with the prior art, the beneficial effects of the present invention are as follows: Traditional chromatic dispersion (CD) compensation only relies on fixed modules and cannot dynamically adapt to the complex coupled non-linear effects in the MCF–SDM system, including self-phase modulation (SPM), crosstalk (XT), and polarization mode dispersion (PMD). The present invention replaces the traditional CD compensation module with an LDBP compensation module to simultaneously handle dispersion and non-linear damage phase noise. The dispersion compensation layer of the LDBP compensation module uses a FIR filter to approximate the frequency-domain filter for dispersion compensation, directly processes the signals of each core in the time domain, replaces the traditional frequency-domain Fourier transform operation, and reduces the computational complexity. The non-linear compensation layer of the LDBP compensation module uses the signal amplitudes of all cores and polarization channels to calculate the non-linear phase of each channel, corrects the phase of the signal through an exponential function, directly cancels the non-linear phase noise in the time domain, avoids the complex calculations of frequency-domain transformation, and effectively compensates for the non-linear damage in the MCF-SDM long-distance optical communication system; the filter coefficients of the LDBP compensation module are obtained through data-driven optimization by training, without relying on accurate fiber link parameters, solving the problem of dependence on prior knowledge in traditional DBP. During training, the filter coefficients of other linear compensation modules are frozen to ensure the rapid convergence of the LDBP compensation module and improve the training efficiency. Description of the Drawings

[0029] Figure 1 Schematic diagram of the steps of the self-learning non-linear compensation method for the MCF-SDM long-distance system in the embodiment;

[0030] Figure 2 Structural diagram of the LDBP compensation module and the overall digital signal processing with two cores as an example provided by the embodiment of the present invention;

[0031] Figure 3 Structural diagram of the long-distance coherent optical communication simulation based on weakly coupled four-core optical fiber provided by the embodiment of the present invention;

[0032] Figure 4 Simulation Q-factor diagram at different transmission powers provided by the embodiment of the present invention;

[0033] Figure 5 Simulation Q-factor diagram under different crosstalk provided by the embodiment of the present invention;

[0034] Figure 6 Amplitude response diagram provided by the embodiment of the present invention;

[0035] Figure 7 Distribution diagram of the coefficients of the fifth non-linear layer provided by the embodiment of the present invention;

[0036] Figure 8 Internal structural diagram of a computer device in an embodiment. Detailed Embodiments

[0037] When a signal propagates in a multi-core fiber, the interaction between CD, XT, PM, and nonlinear effects generates complex dynamics, posing a significant challenge to compensating for nonlinear impairments. To this end, the present invention provides a self-learning nonlinear compensation method for MCF-SDM long-distance systems in embodiments, including the following steps:

[0038] Step 1: Perform pre-execution linear compensation using known transmitted data and received data. The linear compensation is carried out by a linear impairment compensation module, which includes a dispersion compensation module and other linear impairment compensation modules. The other linear impairment compensation modules include a phase offset compensation module, a frequency offset compensation module, and a multi-input multi-output linear equalization module. Save the filter coefficients of each linear compensation module calculated during pre-training.

[0039] Step 2: Divide the one-dimensional signal sequence of each channel of the original data into multiple blocks, add adjacent data to the head and tail positions of each block, map them into a two-dimensional structure, and add the channel dimension to form three-dimensional input data.

[0040] Step 3: Construct an LDBP compensation module for dispersion compensation. The LDBP compensation module includes several alternately arranged dispersion compensation layers and nonlinear compensation layers. The dispersion compensation layer performs dispersion compensation using a complex FIR filter in the time domain, processing the signals of each core in the time domain separately. The nonlinear compensation layer calculates the nonlinear phase to be compensated for each polarization channel of each fiber core using the amplitudes of the signals from all fiber cores and performs nonlinear phase compensation in the time domain through an exponential function.

[0041] Step 4: Replace the dispersion compensation module with the LDBP compensation module, send the three-dimensional input data formed in Step 2 into the LDBP compensation module and other linear impairment compensation modules other than the dispersion compensation module for training, freeze the filter coefficients in the other linear impairment compensation modules, and use the gradient descent method to optimize the coefficients of the LDBP compensation module until a trained LDBP compensation module is obtained.

[0042] Step 5: Sequentially input the data to be processed into the trained LDBP compensation module and other linear impairment compensation modules other than the dispersion compensation module for impairment compensation. Among them, freeze the filter coefficients of the trained LDBP compensation module and adaptively update the filter coefficients of the other linear impairment compensation modules.

[0043] Specifically, in an embodiment of the present invention, the self-learning nonlinear compensation method for MCF-SDM long-distance systems in the embodiment is divided into two stages: a pre-training stage and a formal training stage. The same training data pair is used for training in both stages. The complete digital signal processing process is to perform CD compensation and other linear impairment compensations sequentially.

[0044] Step 1 is the pre-training stage. Using the known transmitted data and received data, linear compensation is pre-executed through the linear impairment compensation module, and the filter coefficients of each linear impairment compensation module calculated during pre-training are saved. The optimization of each linear impairment compensation module during pre-training is performed with reference to the prior art. The linear impairment compensation module includes a dispersion compensation module and other linear compensation modules. The other linear compensation modules include a phase offset compensation module, a frequency offset compensation module, and a multi-input multi-output linear equalization module.

[0045] After the pre-training stage is completed, Step 2 is executed. The one-dimensional signal sequence of each channel of the original data is divided into multiple blocks, and adjacent data is added to the head and tail positions of each block. For the first block and the last block, additional zero elements are added at the starting position of the first block and the ending position of the last block respectively to align the data, and then it is mapped into a two-dimensional structure, which is represented by the following matrix:

[0046]

[0047] Among them, one row in the two-dimensional matrix corresponds to one block, the number of columns in the two-dimensional matrix represents the amount of data within the block, b is the size of each block, and nb is the total length of the one-dimensional signal sequence data; s is the length of the added additional adjacent data; k is the channel number, representing the core or polarization.

[0048] Subsequently, the channel dimension is added to form three-dimensional input data. The channel dimension includes the core and polarization, and the polarization and the core are treated as equivalent channels for parallel processing. The three-dimensional data structure formed through data preprocessing can be directly fed into the LDBP compensation module in Step 3, and the information loss during neural network training is reduced. The input layer of the LDBP compensation module is a three-dimensional matrix, and the three dimensions of the matrix are the partition block dimension, the channel dimension including the core and polarization, and the time dimension, which retains the spatial correlation and time continuity of each channel in the multi-core optical fiber. The three-dimensional data structure supports parallel processing of all core and polarization signals, efficiently models the crosstalk (XT) and polarization mode dispersion (PMD) between different channels, and can solve the problem that traditional single-channel LDBP cannot handle multi-core interaction.

[0049] In the embodiment, through the setting of Step 2, the overlapping adjacent data between blocks can retain the time continuity of the signal and avoid the boundary information loss caused by block division; the channel dimension includes the core and polarization, supports multi-channel parallel processing, can accurately capture the crosstalk and non-linear interaction between different cores and polarization states, provides a structured input for the LDBP compensation module, improves the fitting ability for complex impairments of the MCF-SDM system, and reduces the information loss during training.

[0050] In the embodiment, in step 3, an LDBP compensation module is constructed. The LDBP compensation module is used for dispersion compensation. The LDBP compensation module includes a plurality of alternately arranged dispersion compensation layers and nonlinear compensation layers. The dispersion compensation layer performs dispersion compensation using a complex FIR filter in the time domain, and processes the signals of each core in the time domain. The amplitude of the signal with polarization x or polarization y in core k after propagating a distance h in the time domain is expressed in the following form:

[0051] y k,x / y (h,t) = F -1 [H(h) × F(u k,x / y (0,t))]

[0052] where H(h) represents the transfer function in the frequency domain, F -1 represents the Fourier transform, t represents time, and F(u k,x / y (0,t)) represents the frequency domain signal obtained by performing a Fourier transform on the initial time domain signal with polarization x or y and propagation distance 0 in core k; Dispersion compensation is effectively achieved through the frequency domain compensation operator H -1 (h). The dispersion compensation layer uses a complex FIR filter in the time domain to approximate the frequency domain compensation operator H -1 (h) for dispersion compensation, and processes the signals of each core in the time domain. The output of the dispersion compensation layer is shown in the following formula:

[0053]

[0054] where x x∣yk (i) represents the uncompensated original signal with polarization x or y in core k, z x∣yk (i) represents the output of the dispersion compensation layer with polarization x or y in core k, and i is the signal sequence number; w k is the weight of the FIR filter, and S is the width of the FIR filter.

[0055] In implementation, the dispersion compensation layer directly realizes the inverse operation in the frequency domain through the FIR filter in the time domain, avoiding the Fourier transform, greatly reducing the calculation amount, adapting to the high-speed real-time communication scenario. The filter coefficients are optimized through data-driven, and adaptively fit the frequency domain compensation operator H -1 (h), and do not need to rely on accurate optical fiber link parameters, which can solve the problem of dependence on prior knowledge of traditional DBP.

[0056] The nonlinear compensation layer calculates the nonlinear phase to be compensated for each polarization channel of each optical fiber core using the amplitudes of the signals from all optical fiber cores, and performs nonlinear phase compensation in the time domain through an exponential function. The calculated nonlinear phase is shown in the following formula:

[0057] f(z x∣yk (i)) = exp[-iΦx∣yk (i)]z x∣yk (i)

[0058]

[0059] where z x∣yk (i) represents the output of the dispersion compensation layer with polarization x or polarization y in core k, α xkp and α ykp are trainable parameters, i is the signal number, C is the total number of cores; p is the core number, and core p is the number of another core interacting with core k, with a range of [1, C].

[0060] In the embodiment, the nonlinear compensation layer uses the signal amplitudes of all cores and polarization channels to calculate the nonlinear phase of each channel, corrects the phase of the signal through an exponential function, directly cancels the nonlinear phase noise in the time domain, and avoids the complex calculations of frequency domain transformation.

[0061] In step 4 of the embodiment, the LDBP compensation module is used to replace the dispersion compensation module, and the three-dimensional input data formed in step 2 is sent to the LDBP compensation module and other linear impairment compensation modules other than the dispersion compensation module for training. The filter coefficients in other linear impairment compensation modules are frozen to ensure the rapid convergence of the LDBP compensation module and improve the training efficiency; the gradient descent method is used to optimize the coefficients of the LDBP compensation module until a trained LDBP compensation module is obtained. When using the gradient descent method to optimize the coefficients of the LDBP compensation module, the error between the center b data of each channel in the output of the last linear compensation filter in other linear impairment compensation modules and the known transmitted signal for training is used to update all the coefficients in the LDBP compensation module through gradient backpropagation. The loss function is:

[0062]

[0063] where L MSE is the minimum mean square error value, H xij is the signal output from the last linear compensation filter after the output of the LDBP compensation module is input to other linear compensation modules, T xij is the known transmitted data, and C is the total number of cores.

[0064] In step 5, the training phase has been completed, and the full signal processing phase is entered. The data to be processed is sequentially input into other linear impairment compensation modules except the trained LDBP compensation module and the dispersion compensation module to perform impairment compensation on the data, including linear impairments and non-linear impairments. Among them, the filter coefficients of the trained LDBP compensation module are frozen and no longer updated, and the filter coefficients of other linear impairment compensation modules are adaptively updated, and other linear impairment compensation modules.

[0065] The method provided in the embodiment breaks through the limitation of the traditional method that only targets a single spatial channel, fully considers the impact of crosstalk between different spatial channels on the non-linear effect, is more applicable to the MCF-SDM system, and makes up for the deficiencies of the prior art in dealing with non-linear compensation in multi-core fiber communication systems.

[0066] To further verify the performance of the present invention, a channel model and the split-step Fourier method based on a weakly coupled MCF-SDM system are used in the implementation. A weakly coupled four-core fiber long-distance coherent optical communication simulation system as shown in Figure 3 is built. A four-channel dual-polarization 10-Gbaud QPSK signal is transmitted through a 4-core fiber in 10 spans of 100 km each, with a carrier frequency of 193.4 THz, a total of 8 channels, and each fiber core is amplified separately. At the receiver, the signal is sampled at twice the symbol rate. The attenuation coefficient of the fiber is 0.2 dB / km, the dispersion coefficient is 18.8 ps / nm / km, and the non-linear coefficient is 0.5×10 -3 W / m. The XT range is from -76.6 dB / km to -46.6 dB / km. The transmitter uses a Gaussian filter with a 3-dB bandwidth of 15 GHz. The noise figure of the optical amplifier in each span is 6 dB. Phase noise and frequency offset noise are not introduced in the simulation. The first 6016 symbols of each channel are used for training, and the subsequent 10,368 symbols are used for testing. The width of the dispersion compensation filter is 71. Each block of each channel consists of 128 samples, with 60 adjacent samples added at both ends.

[0067] In the embodiment, two baseline methods are compared: the traditional DBP with manually brute-force optimized coefficients and the LDBP that only considers SPM and does not consider XT (labeled as LDPP(SPM) in the figure). Figure 4The Q-factor performance at different launch powers when XT is 76.6 dB / km. Compared with CD compensation, DBP with a step size per span (StPS) of 1 (1StPS-DBP), 3StPS-DBP, and LDBP (SPM), the proposed LDBP (labeled as LDBP(SPM+XT) in the figure) not only achieves Q-factor gains of 0.25 dB, 0.17 dB, 0.17 dB, and 0 dB respectively at the optimal launch power, but also shows a considerable SNR gain of approximately 6.4 dB, 3.9 dB, 3.4 dB, and 1.7 dB. Figure 5 The results of nonlinear compensation at a launch power of 14 dBm under different XT levels, which confirm the superior performance of the proposed LDBP in dealing with strong nonlinear effects.

[0068] Figure 6 The frequency-domain amplitude response when XT is -76.6 dB / km and the launch power is 14 dBm for five CD compensation layers is shown. The amplitude response is in an "M" shape. The amplitude responses of the first layer and the fifth layer are similar and relatively flat, while the middle three layers show large fluctuations, which can provide a useful reference for manually adjusting the parameters of the traditional DBP for weakly coupled MCF-SDM systems. Figure 7 The normalized nonlinear phase coefficient of the nonlinear compensation layer in the fifth layer in this case is shown. It can be seen that the coefficients of the two polarizations of the same core are relatively large but slightly different, while the coefficients of other cores are small but not negligible, verifying the importance of considering different polarization coefficients for different cores in this method.

[0069] Verified by simulation experiments, compared with the traditional DBP and the LDBP that only considers SPM without considering XT, the scheme provided in the embodiment achieves a higher Q-factor gain and a considerable SNR gain at the optimal launch power, and shows superior performance in dealing with strong nonlinear effects, providing a more effective solution for the nonlinear compensation of MCF-SDM long-haul optical communication systems.

[0070] In an embodiment of the present invention, a computer device is also provided, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the self-learning nonlinear compensation method for MCF-SDM long-haul optical communication systems as described above.

[0071] This computer device can be a terminal, and its internal structure diagram can be as Figure 8As shown. The computer device includes a processor, a memory, a network interface, a display screen, and an input device connected by a bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it implements a self-learning MCF-SDM long-distance optical communication system nonlinear compensation method. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer covering the display screen, or a button, a trackball, or a touchpad provided on the housing of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0072] The memory can be, but is not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), Electric Erasable Programmable Read-Only Memory (EEPROM), etc. Among them, the memory is used to store programs, and the processor executes the programs after receiving execution instructions.

[0073] The processor can be an integrated circuit chip with the ability to process signals. The above-mentioned processor can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc. The processor can also be other general-purpose processors, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. It can implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of the present application. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc.

[0074] Those skilled in the art can understand that Figure 8 the structure shown in [the figure] is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0075] In an embodiment of the present invention, there is also provided a computer-readable storage medium, on which a program is stored. When the program is executed by a processor, it implements the self-learning MCF-SDM long-distance optical communication system nonlinear compensation method as described above.

[0076] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a computer device, or a computer program product. Therefore, the embodiments of the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the embodiments of the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0077] Embodiments of the present invention are described with reference to the flowcharts and / or block diagrams of methods, computer devices, or computer program products according to embodiments of the present invention. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing terminal devices generate a device for realizing the functions specified in the flowchart and / or block diagram.

[0078] These computer program instructions can also be stored in a computer-readable memory capable of guiding the computer or other programmable data processing terminal devices to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device, and the instruction device realizes the functions specified in the flowchart.

[0079] In an embodiment of the present invention, a computer program product is further provided, including a computer program / instructions, and when the computer program / instructions are executed by a processor, the steps of the above method are realized.

[0080] In actual application processes, the above computer program product includes but is not limited to: smart phones, desktop computers, laptop computers, tablet computers, host computers, and server platforms, etc., and no specific limitations are made here.

[0081] The above has introduced in detail the application of the self-learning MCF-SDM long-distance optical communication system nonlinear compensation method, system, computer device, computer-readable storage medium, and computer program product provided by the present invention. Specific examples are used in this article to elaborate on the principle and implementation manner of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A self-learning MCF-SDM long-distance system nonlinear compensation method, characterized in that: The following steps are involved: Step 1: Using known transmission data and reception data, pre-perform linear compensation through a linear damage compensation module, and save filter coefficients of each linear damage compensation module calculated during pre-training, wherein the linear damage compensation module includes a dispersion compensation module, a phase offset compensation module, a frequency offset compensation module, and a multi-input multi-output linear equalization module; Step 2: Divide the one-dimensional signal sequence of each channel of the original data into multiple blocks, add adjacent data to the head and tail positions of each block and map them into a two-dimensional structure, add the channel dimension, and form three-dimensional input data; Step 3: constructing an LDBP compensation module, wherein the LDBP compensation module is used for dispersion compensation, and the LDBP compensation module comprises a plurality of dispersion compensation layers and nonlinear compensation layers arranged alternately; Step 4: Using the LDBP compensation module to replace the dispersion compensation module, the three-dimensional input data formed in step 2 is sent to the LDBP compensation module and other linear damage compensation modules other than the dispersion compensation module for training, the filter coefficients in other linear damage compensation modules are frozen, and the coefficients of the LDBP compensation module are optimized using the gradient descent method until a trained LDBP compensation module is obtained; Step 5: sequentially input the data to be processed into the trained LDBP compensation module and other linear damage compensation modules other than the dispersion compensation module to perform damage compensation on the data, wherein the filter coefficients of the trained LDBP compensation module are frozen, and the filter coefficients of other linear damage compensation modules are adaptively updated.

2. The self-learning MCF-SDM long-distance system nonlinear compensation method according to claim 1, characterized in that: In step 2, when adding adjacent data to the first and last positions of each block, for the first block and the last block, additional zero elements are added to the start position of the first block and the end position of the last block respectively to align the data.

3. The self-learning MCF-SDM long-distance system nonlinear compensation method according to claim 2, characterized in that: In step 2, the two-dimensional structure of the data is represented by the following matrix: Among them, a row in the two-dimensional matrix corresponds to a block, the number of columns in the two-dimensional matrix represents the amount of data in the block, b is the size of each block, nb is the total length of the one-dimensional signal sequence data; s is the length of the added additional adjacent data; k is the channel number, indicating the core or polarization; plus the channel dimension, three-dimensional input data is formed. The channel dimension includes the core and polarization. Polarization and core are treated as equivalent channels and processed in parallel.

4. The self-learning MCF-SDM long-distance system nonlinear compensation method according to claim 1, characterized in that: The dispersion compensation layer uses a complex FIR filter to perform dispersion compensation in the time domain, and processes the signal of each fiber core in the time domain separately. The nonlinear compensation layer uses the amplitude of the signals from all fiber cores to calculate the nonlinear phase that needs to be compensated for each polarization channel of each fiber core, and performs nonlinear phase compensation in the time domain through an exponential function.

5. The self-learning MCF-SDM long-distance system nonlinear compensation method according to claim 4, characterized in that: In step 3, the amplitude of the signal with polarization x or polarization y in the fiber core k after propagating a distance h in the time domain is expressed in the following form: u k,x / y (h,t)=F -1 [H(h)×F(u k,x / y (0,t))] Where H(h) represents the transfer function in the frequency domain, F -1 represents Fourier transform, t represents time, F(u k,x / y (0,t)) represents the frequency domain signal obtained by Fourier transforming the initial time domain signal when the propagation distance is 0 and the polarization is x or y in the fiber core k; -1 (h) Effectively realize dispersion compensation. The dispersion compensation layer uses a complex FIR filter in the time domain to approximate the frequency domain compensation operator H -1 (h) Dispersion compensation is performed, and the signal of each fiber core in the time domain is processed separately. The output of the dispersion compensation layer is shown as follows: where x x∣yk (i) represents the original uncompensated signal of polarization x or polarization y in core k, z x∣yk (i) represents the output of the dispersion compensation layer of polarization x or polarization y in the core k, i is the signal sequence number; w k is the weight of the FIR filter, and S is the width of the FIR filter.

6. The self-learning MCF-SDM long-distance system nonlinear compensation method according to claim 5, characterized in that: In step 3, the nonlinear compensation layer uses the amplitude of the signal from all the optical fiber cores to calculate the nonlinear phase that needs to be compensated for each polarization channel of each core, and directly performs nonlinear phase compensation in the time domain through an exponential function. The calculated nonlinear phase is shown in the following formula: f(z x∣yk (i))=exp[-iΦ x∣yk (i)]z x∣yk (i) where z x∣yk (i) represents the output of the dispersion compensation layer for polarization x or polarization y in core k, α xkp and α ykp is a trainable parameter, i is the signal number, C is the total number of cores; p is the core number, core p is the number of another core that interacts with core k, and the range is [1, C].

7. The self-learning MCF-SDM long-distance system nonlinear compensation method according to claim 1, characterized in that: In step 4, the coefficients of the LDBP compensation module are optimized using the gradient descent method. The error between the center b data of each channel in the output of the last linear compensation filter in other linear damage compensation modules and the known transmission signal used for training is used to update all coefficients in the LDBP compensation module through gradient back propagation. The loss function is: Where L MSE is the minimum mean square error value, Hx ij The output of the LDBP compensation module is input to the last linear compensation filter output signal after the other linear compensation modules, Tx ij is the known transmitted data, and C is the total number of fiber cores.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the self-learning MCF-SDM long-distance optical communication system nonlinear compensation method as claimed in claim 1 is implemented.

9. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, the nonlinear compensation method of the self-learning MCF-SDM long-distance optical communication system as claimed in claim 1 is implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method according to claim 1 are implemented.

Images