Training method of end-to-end inter-symbol interference compensation model, interference compensation method and equipment
By optimizing the tap coefficients of THP and FFE through an end-to-end training method, the problems of inter-symbol interference and other impairments in the super Nyquist coherent optical transmission system were solved, thereby improving the system performance.
Patent Information
- Application Number
- CN202511452168.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-11
- Publication Date
- 2026-02-13
AI Technical Summary
In existing technologies for super Nyquist coherent optical transmission systems, the traditional minimum mean square error (MINS) tap coefficient is difficult to achieve global optimization in complex transmission environments, and cannot effectively compensate for inter-symbol interference and other impairments, resulting in limited system performance.
The training method of the end-to-end inter-symbol interference compensation model is adopted, which treats the transmitting and receiving processing modules as a whole system. By iteratively optimizing the tap coefficients of THP and FFE, direct compensation for inter-symbol interference is achieved. The gradient pass-through estimation method is used for model training to avoid dependence on precise mathematical modeling of the channel.
It significantly reduces the system bit error rate, improves the optical signal-to-noise ratio tolerance, and increases the transmission capacity and distance of the ultra-Nyquist coherent optical transmission system.
Smart Images
Figure CN121530484A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of optical communication, and in particular to a training method for an end-to-end inter-symbol interference compensation model, an inter-symbol interference compensation method, and a device. Background Technology
[0002] With the explosive growth of global Internet Protocol traffic and data center interconnect services, the demand for higher spectral efficiency in coherent optical transmission systems has become an inevitable trend in the industry. However, the increase in throughput of coherent optical transmission systems is fundamentally limited by the available bandwidth of core optoelectronic devices such as digital-to-analog converters (DACs) and analog-to-digital converters (ADCs). To overcome this bottleneck, super Nyquist signaling technology has emerged. This technology actively compresses the bandwidth of the transmitted signal, introducing strong but controllable inter-symbol interference into the signal, thereby achieving higher data transmission rates and spectral efficiency in exchange for higher data transmission rates, providing a promising solution for improving system capacity.
[0003] Despite the significant advantages of Super Nyquist signaling, the strong inter-symbol interference (ISI) it introduces must be effectively compensated for using complex digital signal processing techniques. Tomlinson-Harashima precoding (THP), which processes the signal sequence through feedback filters and modulo operations, has proven highly effective in mitigating ISI. In traditional schemes, the THP tap coefficients are typically calculated at the receiver using a minimum mean square error (MSE) decision feedback equalizer, and then the coefficients of the feedback equalization (FBE) portion are transferred to the transmitter, while the feedforward equalization (FFE) portion remains at the receiver. However, in Super Nyquist coherent optical transmission systems, in addition to strong ISI, there are also multiple complex impairments such as polarization mode dispersion, fiber nonlinearity, and optoelectronic device nonlinearity. Therefore, whether this traditional MSE-THP tap coefficient approach can achieve global optimization of the entire Super Nyquist coherent optical transmission system under such complex transmission conditions is uncertain. Thus, a method that can directly optimize the system globally to effectively compensate for ISI is urgently needed. Summary of the Invention
[0004] In view of this, embodiments of this application provide a method and apparatus for determining event trigger words, so as to eliminate or improve one or more defects existing in the prior art.
[0005] This application provides a training method for an end-to-end inter-symbol interference compensation model, the method comprising: In the current iteration, the training data sequence is input into the current end-to-end system model used to simulate the super Nyquist coherent optical transmission system. This enables the transmitter processing module in the end-to-end system model to generate the simulated transmission signal corresponding to the training data sequence, and the channel module in the end-to-end system model to transmit the simulated transmission signal in a preset channel environment to obtain the received complex symbol sequence. The receiver processing module in the end-to-end system model then performs inter-symbol interference compensation on the received complex symbol sequence and outputs the interference compensation result data corresponding to the training data sequence. The target loss for the current iteration is determined based on the interference compensation result data, and the model parameters of the transmitter processing module and the receiver processing module are optimized based on the loss to update the end-to-end system model. If the end-to-end system model has converged or the current iteration round is the preset last iteration round, then the transmitter processing module and the receiver processing module in the updated end-to-end system model are determined as an inter-symbol interference compensation model for compensating for inter-symbol interference in the received data sequence.
[0006] In some embodiments of this application, the transmitter processing module includes: The mapping module is used to perform QAM mapping on the pseudo-random code used as the training data sequence to obtain the mapping result data; The THP training module is used to pre-encode the mapping result data to obtain a pre-encoded complex symbol sequence; The Super Nyquist shaping module is used to shape the pre-encoded complex symbol sequence to obtain the simulated transmission signal.
[0007] In some embodiments of this application, the THP training module includes: The first one-dimensional convolutional layer is used to filter the mapping result data to obtain the filtered first complex symbol sequence. The convolution kernel coefficients of the first one-dimensional convolutional layer are the tap coefficients to be trained in the THP training module. The first modulo operation unit is used to perform modulo operation on the filtered first complex number symbol sequence to generate a pre-encoded complex number symbol sequence.
[0008] In some embodiments of this application, the super Nyquist shaping module includes: An upsampling unit is used to perform upsampling processing on the pre-encoded complex symbol sequence to obtain upsampling result data; The super Nyquist compression unit is used to perform super Nyquist compression on the upsampled result data to obtain compressed result data; A root-raised cosine filter is used to pulse-shape the compressed data to obtain a simulated transmission signal.
[0009] In some embodiments of this application, the receiving end processing module includes an FFE training module, which includes: The second one-dimensional convolutional layer is used to filter the received complex symbol sequence to obtain the filtered second complex symbol sequence. The convolution kernel coefficients of the second one-dimensional convolutional layer are the tap coefficients to be trained in the FFE training module. The second modulo operation unit is used to perform modulo operation on the filtered second complex number symbol sequence to generate interference compensation result data.
[0010] In some embodiments of this application, determining the target loss for the current iteration based on the interference compensation result data, and optimizing the model parameters of the transmitting end processing module and the receiving end processing module based on the loss to update the end-to-end system model, includes: The mean square error between the interference compensation result data and the mapping result data is calculated. The mean square error is used as the loss function value of the end-to-end system model corresponding to the current iteration round. Based on the gradient optimization algorithm, the gradient of the model parameters is calculated according to the loss function value, and the model parameters are updated using the gradient. The model parameters include the tap coefficients of the THP training module and the tap coefficients of the FFE training module.
[0011] Another aspect of this application provides a first end-to-end inter-symbol interference compensation method, comprising: The target data sequence is input into the transmitter processing module so that the transmitter processing module outputs the target transmission signal corresponding to the target data sequence. The target transmission signal is transmitted to the receiving end, so that the receiving end receives the complex symbol sequence corresponding to the target transmission signal and inputs the complex symbol sequence into the receiving end processing module, thereby enabling the receiving end to obtain the interference compensation result data corresponding to the complex symbol sequence output by the receiving end processing module. The transmitting end processing module and the receiving end processing module constitute an inter-symbol interference compensation model for compensating for inter-symbol interference in the received data sequence. This inter-symbol interference compensation model is pre-trained based on the training method of the end-to-end inter-symbol interference compensation model.
[0012] A third aspect of this application provides a second end-to-end inter-symbol interference compensation method, comprising: The receiver receives a complex symbol sequence corresponding to a target transmission signal transmitted over a network and inputs the complex symbol sequence into a receiver processing module, so that the receiver processing module outputs interference compensation result data corresponding to the complex symbol sequence; wherein, the target transmission signal is output by the transmitter processing module after the transmitter has input a target data sequence into the transmitter processing module in advance. The transmitting end processing module and the receiving end processing module constitute an inter-symbol interference compensation model for compensating for inter-symbol interference in the received data sequence. This inter-symbol interference compensation model is pre-trained based on the training method of the end-to-end inter-symbol interference compensation model.
[0013] A fourth aspect of this application provides an electronic device including a processor and a memory, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement at least one of the training method for the end-to-end inter-symbol interference compensation model, the first end-to-end inter-symbol interference compensation method, and the second end-to-end inter-symbol interference compensation method.
[0014] The fifth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements at least one of the training method for the end-to-end inter-symbol interference compensation model, the first end-to-end inter-symbol interference compensation method, and the second end-to-end inter-symbol interference compensation method.
[0015] The sixth aspect of this application provides a computer program product comprising a computer program that, when executed by a processor, implements at least one of the training method for the end-to-end inter-symbol interference compensation model, the first end-to-end inter-symbol interference compensation method, and the second end-to-end inter-symbol interference compensation method.
[0016] The training method for the end-to-end inter-symbol interference (ISI) compensation model provided in this application treats the transmitter, channel, and receiver as a unified end-to-end system for training. In the current iteration, the training data sequence is input into the current end-to-end system model, enabling the transmitter processing module to generate a simulated transmission signal and the channel module to transmit the simulated transmission signal to obtain a received complex symbol sequence. The receiver processing module then performs ISI compensation on the received complex symbol sequence and outputs the interference compensation result data. Based on the interference compensation result data, the target loss for the current iteration is determined, and the model parameters are optimized to update the model. If the model has converged, the updated transmitter and receiver processing modules are used as the model for ISI compensation of the received data sequence. This application enables joint optimization of the transmitter and receiver processing modules, thereby effectively compensating for ISI, significantly reducing the system bit error rate, improving the optical signal-to-noise ratio (SNR) tolerance, and ultimately increasing the transmission capacity and distance of the entire super Nyquist coherent optical transmission system.
[0017] Additional advantages, objectives, and features of this application will be set forth in part in the description which follows, and will in part become apparent to those skilled in the art upon review of the following description, or may be learned by practice of the application. The objectives and other advantages of this application can be realized and obtained by means of the structures specifically pointed out in the specification and drawings.
[0018] Those skilled in the art will understand that the purposes and advantages that can be achieved with this application are not limited to those specifically described above, and that the above and other purposes that this application can achieve will be more clearly understood from the following detailed description. Attached Figure Description
[0019] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, are not intended to limit the scope of the application. The components in the drawings are not drawn to scale but are merely for illustrating the principles of the application. For ease of illustration and description of certain parts of the application, corresponding portions in the drawings may be enlarged, i.e., may appear larger relative to other components in an exemplary device actually manufactured according to this application. In the drawings: Figure 1 This is a schematic diagram of the first step in the training method of the end-to-end inter-symbol interference compensation model in one embodiment of this application.
[0020] Figure 2 This is a schematic diagram of the structure of an end-to-end inter-symbol interference compensation model in one embodiment of this application.
[0021] Figure 3 This is a flowchart illustrating step 200 in the training method of the end-to-end inter-symbol interference compensation model in one embodiment of this application.
[0022] Figure 4 This is a second flowchart illustrating the training method of the end-to-end inter-symbol interference compensation model in one embodiment of this application.
[0023] Figure 5 This is a schematic diagram illustrating the principle of the training method for the end-to-end inter-symbol interference compensation model in one example of this application.
[0024] Figure 6 This is a comparison diagram in the time domain of the signal transmitted through the real channel and the signal generated by the CGAN channel module, which is the training method of the end-to-end inter-symbol interference compensation model in one embodiment of this application.
[0025] Figure 7(a) is a comparison of the first normalized amplitude response of THP and FFE in the frequency domain obtained by the training method of the end-to-end inter-symbol interference compensation model in an embodiment of this application with that obtained by the conventional method.
[0026] Figure 7(b) is a comparison of the second normalized amplitude response of THP and FFE in the frequency domain obtained by the training method of the end-to-end inter-symbol interference compensation model in one embodiment of this application with that obtained by the conventional method.
[0027] Figure 8 This is a graph showing the comparison of bit error rate performance between the training method of the end-to-end inter-symbol interference compensation model in one embodiment of this application and traditional methods under different optical signal-to-noise ratios.
[0028] It should also be noted that, in order to avoid obscuring this application with unnecessary details, only the processing steps closely related to the solution according to this application are shown in the accompanying drawings, while other details that are not closely related to this application are omitted. Detailed Implementation
[0029] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the embodiments and accompanying drawings. Here, the illustrative embodiments and their descriptions are used to explain this application, but are not intended to limit it.
[0030] It should also be noted that, in order to avoid obscuring this application with unnecessary details, only the structures and / or processing steps closely related to the solution according to this application are shown in the accompanying drawings, while other details that are not closely related to this application are omitted.
[0031] It should be emphasized that the term "including / comprises" as used herein refers to the presence of a feature, element, step, or component, but does not exclude the presence or addition of one or more other features, elements, steps, or components.
[0032] It should also be noted that, unless otherwise specified, the term "connection" in this article can refer not only to a direct connection, but also to an indirect connection involving an intermediary.
[0033] In the following description, embodiments of the present application will be illustrated with reference to the accompanying drawings. In the drawings, the same reference numerals represent the same or similar parts, or the same or similar steps.
[0034] The following examples will provide a detailed description.
[0035] This application provides a training method for an end-to-end inter-symbol interference compensation model, see [link to relevant documentation]. Figure 1 The training method for the end-to-end symbol interference compensation model specifically includes the following: Step 100: In the current iteration round, the training data sequence is input into the current end-to-end system model, so that the transmitter processing module in the end-to-end system model generates the simulated transmission signal corresponding to the training data sequence, and the channel module in the end-to-end system model transmits the simulated transmission signal in a preset channel environment to obtain the received complex symbol sequence, and the receiver processing module in the end-to-end system model performs inter-symbol interference compensation on the received complex symbol sequence and outputs the interference compensation result data corresponding to the training data sequence.
[0036] In one or more embodiments of this application, the training data sequence can generate two independent pseudo-random binary sequences for the system. The complex symbol sequence refers to a series of complex ordered sets representing signal amplitude and phase information. The simulated transmission signal refers to the complex symbol sequence obtained by the transmitter processing module after processing the training data sequence at the transmitter. The preset channel environment refers to the complete optical path transmission environment from the output of the simulated transmission signal simulated by the channel module to the input of the receiver processing module. The received complex symbol sequence refers to the complex symbol sequence after transmission through the channel. The interference compensation result data refers to the complex symbol sequence after inter-symbol interference compensation by the receiver processing module.
[0037] In step 100, see Figure 2The end-to-end system model includes a transmitter processing module, a channel module, and a receiver processing module. The transmitter processing module A100 performs transmitter simulation processing on the training data sequence to output a simulated transmission signal. The channel module A200 transmits the simulated transmission signal under a preset channel environment to output a received complex symbol sequence. The receiver processing module A300 performs receiver processing on the received complex symbol sequence to output interference compensation result data. The iterative processing module A400 determines the target loss of the current iteration round based on the interference compensation result data and optimizes the model parameters of the transmitter processing module and the receiver processing module based on the loss to update the end-to-end system model. If the end-to-end system model has converged or the current iteration round is the preset last iteration round, then the transmitter processing module and the receiver processing module in the updated end-to-end system model are determined as an inter-symbol interference compensation model for compensating for inter-symbol interference in the received data sequence.
[0038] Step 200: Determine the target loss of the current iteration based on the interference compensation result data, and optimize the model parameters of the transmitter processing module and the receiver processing module based on the target loss to update the end-to-end system model.
[0039] In one or more embodiments of this application, the target loss can be the mean square error between two complex symbol sequences, and the model parameters refer to the tap coefficients used to adjust the signal amplitude and phase to compensate for channel impairments.
[0040] Step 300: If the end-to-end system model has converged or the current iteration round is the preset last iteration round, then the transmitter processing module and the receiver processing module in the updated end-to-end system model are determined as an inter-symbol interference compensation model for compensating for inter-symbol interference of the received data sequence.
[0041] It should be noted that, in one or more embodiments of this application, the transmitting end processing module and the receiving end processing module can be used separately during the model application stage.
[0042] In step 300, it can be ensured that the end-to-end system model is updated at least once, and the end-to-end system model updated in the current iteration is used as the initial end-to-end system model for the next iteration.
[0043] As can be seen from the above description, the training method of the end-to-end inter-symbol interference compensation model provided in this application treats the entire communication link, including the transmitter, channel and receiver, as a trainable end-to-end system model. By jointly optimizing the precoder THP of the transmitter and the feedforward equalizer FFE of the receiver, effective compensation for inter-symbol interference is achieved, thereby significantly improving the system transmission performance.
[0044] To further achieve effective compensation for inter-symbol interference, a training method for an end-to-end inter-symbol interference compensation model is provided in this application embodiment, see [link to relevant documentation]. Figure 2 The transmitter processing module A100 in the end-to-end inter-symbol interference compensation model in step 100 specifically includes: Mapping module A110: used to perform QAM mapping on the pseudo-random code used as the training data sequence to obtain the mapping result data; THP training module A120 is used to pre-encode the mapping result data to obtain a pre-encoded complex symbol sequence; The Super Nyquist Shaping Module A130 is used to shape the pre-encoded complex symbol sequence to obtain a simulated transmission signal.
[0045] Correspondingly, the specific data processing process for generating the simulated transmission signal corresponding to the training data sequence by the transmitter processing module in the end-to-end system model can be as follows: The pseudo-random code used as the training data sequence is input into the mapping module in the transmitter processing module, so that the mapping module performs QAM mapping on the pseudo-random code used as the training data sequence to obtain the mapping result data, which is then transmitted to the THP training module and the iterative processing module in the end-to-end symbol interference compensation model. The THP training module then pre-encodes the mapping result data to obtain a pre-encoded complex symbol sequence, which is then transmitted to the super Nyquist shaping module. The super Nyquist shaping module then shapes the pre-encoded complex symbol sequence to obtain the simulated transmission signal, which is then transmitted to the channel module in the end-to-end symbol interference compensation model.
[0046] It is understandable that the mapping result data refers to the complex symbol sequence formed by modulating the pseudo-random code of the training data sequence to be transmitted through the mapping module, and the pre-encoded complex symbol sequence refers to the complex symbol sequence pre-encoded by the THP training module.
[0047] To further achieve effective compensation for inter-symbol interference, a training method for an end-to-end inter-symbol interference compensation model is provided in this application embodiment, see [link to relevant documentation]. Figure 2 The THP training module A120 in the training method of an end-to-end inter-symbol interference compensation model specifically includes: The first one-dimensional convolutional layer A121 is used to filter the mapping result data to obtain the filtered first complex symbol sequence; First modulo operation unit A122: used to perform modulo operation on the filtered first complex number symbol sequence to generate a pre-encoded complex number symbol sequence.
[0048] Correspondingly, the specific data processing process for the THP training module to pre-encode the mapping result data to obtain the pre-encoded complex symbol sequence can be as follows: the mapping result data is input into the first one-dimensional convolutional layer in the THP training module, so that the first one-dimensional convolutional layer filters the mapping result data to obtain the filtered first complex symbol sequence and transmits it to the first modulo operation unit, so that the first modulo operation unit performs modulo operation on the filtered first complex symbol sequence to obtain the pre-encoded complex symbol sequence, and then transmits it to the super Nyquist shaping module in the THP training module.
[0049] It is understandable that the first complex symbol sequence after filtering refers to the complex symbol sequence after being filtered by the first one-dimensional convolutional layer.
[0050] In one or more embodiments of this application, the convolution kernel coefficients of the first one-dimensional convolutional layer can be the tap coefficients to be trained in the THP training module. .
[0051] Specifically, the input mapping result data is first filtered through the first one-dimensional convolutional layer in the feedback loop, and then processed by modulo operation in the first modulo operation unit to generate a pre-encoded complex number symbol sequence. The process expression is as follows: in, The input symbols after mapping. For the pre-encoded output symbols, These are the trainable tap coefficients of the first one-dimensional convolutional layer in the THP training module. The length of the THP tap coefficient. The modulo operation is defined as follows: in, Given the input sequence, The amplitude of the signal is limited to a preset modulo operation boundary value. The power amplification caused by the internal feedback operation is limited.
[0052] To further achieve effective compensation for inter-symbol interference, a training method for an end-to-end inter-symbol interference compensation model is provided in this application embodiment, see [link to relevant documentation]. Figure 2 The super Nyquist shaping module A130 in the end-to-end inter-symbol interference compensation model training method specifically includes: Upsampling unit A131: used to perform upsampling processing on the pre-encoded complex symbol sequence to obtain upsampling result data; Super Nyquist compression unit A132: used to perform super Nyquist compression on the upsampled result data to obtain compressed result data; Root-raised cosine filter A133: used to pulse-shape the compressed result data to obtain the simulated transmission signal.
[0053] Correspondingly, the specific data processing procedure for the precoded complex symbol sequence to be shaped by the Super Nyquist shaping module to obtain the simulated transmission signal, and then transmitted to the channel module in the end-to-end symbol interference compensation model, can be as follows: the precoded complex symbol sequence is input into the upsampling unit in the Super Nyquist shaping module, so that the upsampling unit performs upsampling processing on the precoded complex symbol sequence to obtain upsampled result data and transmits it to the Super Nyquist compression unit, so that the Super Nyquist compression unit performs Super Nyquist compression on the upsampled result data to obtain compressed result data and transmits it to the root-raised cosine filter, so that the root-raised cosine filter performs pulse shaping on the compressed result data to obtain the simulated transmission signal, and then transmits it to the channel module in the end-to-end symbol interference compensation model.
[0054] It is understandable that the upsampling result data refers to the complex symbol sequence after upsampling processing by the upsampling unit, while the compressed result data refers to the complex symbol sequence after being shaped by the super Nyquist shaping module.
[0055] To further achieve effective compensation for inter-symbol interference, a training method for an end-to-end inter-symbol interference compensation model is provided in this application embodiment, see [link to relevant documentation]. Figure 2 The receiver processing module A300 in the end-to-end inter-symbol interference compensation model training method specifically includes: FFE training module A310: used to perform feedforward equalization processing on the received complex symbol sequence to output interference compensation result data.
[0056] Correspondingly, the specific data processing process by which the receiving end processing module in the end-to-end system model performs inter-symbol interference compensation on the received complex symbol sequence and outputs the interference compensation result data corresponding to the training data sequence can be as follows: the received complex symbol sequence is input into the FFE training module in the receiving end processing module, so that the FFE training module performs signal recovery processing on the received complex symbol sequence to obtain the interference compensation result data, and then transmits it to the iterative processing module in the end-to-end symbol interference compensation model.
[0057] The FFE training module A310 specifically includes: The second one-dimensional convolutional layer A311 is used to filter the received complex symbol sequence to obtain a filtered second complex symbol sequence. The second modulo operation unit A312 is used to perform modulo operation on the filtered second complex number symbol sequence to generate interference compensation result data.
[0058] Correspondingly, the specific data processing procedure of inputting the received complex symbol sequence into the FFE training module of the receiving end processing module, so that the FFE training module performs signal recovery processing on the received complex symbol sequence to obtain interference compensation result data, and then transmits it to the iterative processing module in the end-to-end symbol interference compensation model, can be as follows: inputting the received complex symbol sequence into the second one-dimensional convolutional layer in the FFE training module, so that the second one-dimensional convolutional layer filters the received complex symbol sequence to obtain a filtered second complex symbol sequence and transmits it to the second modulo operation unit, so that the second modulo operation unit performs modulo operation on the filtered second complex symbol sequence to obtain interference compensation result data, and then transmits it to the iterative processing module in the end-to-end symbol interference compensation model.
[0059] It is understandable that the filtered second complex symbol sequence refers to the complex symbol sequence after being filtered by the second one-dimensional convolutional layer.
[0060] In one or more embodiments of this application, the convolution kernel coefficients of the second one-dimensional convolutional layer can be the tap coefficients to be trained in the FFE training module. .
[0061] Specifically, the second one-dimensional convolutional layer first filters the input received complex symbol sequence, and then performs modulo operation processing through the second modulo operation unit to generate interference compensation result data. The process expression is as follows: in, The input symbols after mapping. For the pre-encoded output symbols, These are the trainable tap coefficients of the one-dimensional convolutional layer in the FFE training module. The length of the FFE tap coefficient is used here. The same modulo operation as the THP training module is used to solve the problem of the extended signal set caused by the modulation operation at the transmitter.
[0062] To achieve joint optimization of the precoder THP in the transmitter processing module and the feedforward equalizer FFE in the receiver processing module, a training method for an end-to-end inter-symbol interference compensation model is provided in an embodiment of this application, see [link to relevant documentation]. Figure 3 Step 200 specifically includes the following: Step 210: Calculate the mean square error between the interference compensation result data and the mapping result data, and use the mean square error as the loss function value of the end-to-end system model corresponding to the current iteration round.
[0063] Specifically, the interference compensation result data is compared with the mapping result data, and the mean square error between the two is calculated as the loss function of the entire end-to-end model. Its definition expression is as follows: in, The number of samples can be the number of symbols.
[0064] Step 220: Gradient-based optimization algorithm, calculate the gradient of the model parameters based on the loss function value, and update the model parameters using the gradient. The model parameters include the tap coefficients of the THP training module and the tap coefficients of the FFE training module.
[0065] In step 220, since the modulo operation involved in the THP training module and the FFE training module is non-differentiable, it hinders the backpropagation of gradients during the end-to-end system model training process, thus impeding effective model optimization. To solve this problem, embodiments of this application employ a gradient pass-through estimation method. During backpropagation, the gradient pass-through estimation method replaces the non-differentiable operation with an identity function, thereby achieving effective gradient flow and stable model training. After gradient backpropagation is possible, a gradient-based optimization algorithm is used to calculate the tap coefficients of the loss for the THP training module based on the loss function. Tap coefficients of the FFE training module The gradient is calculated, and this gradient information is used to update the value simultaneously. and Repeat this process iteratively until the loss function converges, at which point the result is obtained. and This is the optimal coefficient under the current channel environment.
[0066] It should be noted that although existing technologies have disclosed training methods for inter-symbol interference compensation models, the reliance on precise mathematical modeling and complex channel estimation in traditional techniques leads to performance limitations due to the inaccuracy of traditional models. The embodiments of this application, however, design a training method for inter-symbol interference compensation models. The designers of this application first conceived of learning the optimal compensation strategy by directly minimizing the end-to-end error of the system, implicitly learning the complex transmission characteristics of the channel. This method avoids the reliance on precise mathematical modeling and complex channel estimation in traditional techniques, thus overcoming the performance limitations caused by the inaccuracy of traditional models, and enhancing the robustness and universality of the method in practical applications. The embodiments of this application provide an end-to-end inter-symbol interference compensation model training method that, through a modular approach, allows the channel model in end-to-end training to be replaced according to different channel conditions by simply changing the channel parameters, thereby achieving optimal tuning for different fiber optic channel environments.
[0067] In a specific example, the complete process of training a method for an end-to-end inter-symbol interference compensation model is illustrated as follows: Figure 4 As shown, the training method for the end-to-end inter-symbol interference compensation model is specifically as follows: Based on a super Nyquist dual-polarization 16-orthogonal amplitude-phase modulation (QAM) coherent fiber transmission system, the constructed end-to-end model includes (a) a mapping module, (b) a THP training module, (c) a super Nyquist shaping module, (d) a channel module, and (e) an FFE training module, as follows: Figure 5 As shown. In this embodiment, (d) the channel module is implemented using a pre-trained conditional generative adversarial network (CGAN), which can accurately simulate the complete optical path transmission characteristics from the digital pulse shaping at the transmitter to the FFE training module at the receiver. To ensure that the gradient can pass the modulus operation smoothly during backpropagation, this embodiment uses a pass-through estimator method to process the modulus operation in the THP training module and the FFE training module.
[0068] Specifically, the system generates two independent pseudo-random binary sequences, which are mapped into 16-QAM complex symbol sequences in the mapping module. These 16-QAM complex symbol sequences are then input into the THP training module. In this embodiment, the tap length of the one-dimensional convolutional layer in the THP training module is set to 8, and the boundary value A for the modulo operation is set to 4. The complex symbol sequence processed by the THP training module is then fed into the super Nyquist shaping module, which performs a 4x upsampling and pulse shaping using a root-raised cosine filter with a roll-off factor of 0.1. The super Nyquist compression factor is set to 0.85. The baud rate of the entire system is 30 Gbaud, the bandwidth of the Nyquist signal after pulse shaping is 16.5 GHz, and the actual transmitted signal bandwidth after super Nyquist compression is approximately 14 GHz. The laser center wavelength is 1550 nm, and the linewidth is 100 kHz.
[0069] The signal is transmitted through 80 kilometers of G.652 standard single-mode fiber and erbium-doped fiber amplifiers before being received by a coherent receiver. In this embodiment, the tap length of the one-dimensional convolutional layer of the FFE training module is set to 41. The received complex symbol sequence is input into the FFE training module, and after one-dimensional convolution and modulo operations, the recovered complex symbol sequence is output.
[0070] The complex symbol sequence output by the FFE training module is compared with the original 16QAM complex symbol sequence to calculate the mean squared error loss. This embodiment uses the AdamW optimizer, which iteratively updates the 8 tap coefficients of the THP training module and the 41 tap coefficients of the FFE training module based on the gradient calculated from the loss function. To accelerate convergence, the coefficients of both modules can be initialized to values calculated using the traditional minimum mean squared error algorithm. The training process continues until the loss function converges, marking the completion of end-to-end system model optimization.
[0071] To verify the accuracy of the channel module in this embodiment, such as Figure 6 As shown, the time-domain waveform of the signal after transmission through a real system (e.g.) Figure 6 (shown by the blue dashed line) and the signal waveform generated by the CGAN channel module in this embodiment (as shown by...) Figure 6The results (shown by the red dashed line) show a high degree of agreement, with a normalized mean square error of only 0.016, demonstrating the accuracy of the channel module. To analyze the characteristics of the optimized coefficients, as shown in Figures 7(a) and 7(b), where MMSE-THP represents Minimum Mean Square Error - Tomlinson-Harashima Precoding and E2E-THP represents End-to-End - Tomlinson-Harashima Precoding, compared to the coefficients obtained by the traditional minimum mean square error method (shown by the green curves in Figures 7(a) and 7(b)), the THP obtained through end-to-end training in this embodiment exhibits a deeper notch in the frequency domain response (shown by the red curves in Figures 7(a) and 7(b), more effectively suppressing inter-symbol interference; simultaneously, the end-to-end FFE obtained in this embodiment has higher gain in the high-frequency part, more effectively compensating for the attenuation of high-frequency components caused by super Nyquist compression. To evaluate the overall system performance, as shown in Figures 7(a) and 7(b)... Figure 8 As shown, FTN (τ=0.85) represents the baseline bit error rate (BER) without THP and FFE compensation when transmitting at a super Nyquist compression ratio of 0.85. MMSE-THP and FFE represent minimum mean square error (MSE) - THP and FFE, and E2E-THP and FFE represent end-to-end - THP and FFE. The end-to-end THP and FFE in this embodiment are compared with the THP and FFE using the traditional minimum mean square error method in terms of BER performance. The results show that at the industry-standard 20% soft-decision forward error correction threshold, the method in this embodiment achieves a 1.0 dB improvement in optical signal-to-noise ratio (SNR) compared to the traditional method. This demonstrates that the embodiments of this application, through end-to-end joint optimization, can achieve more effective global compensation for system impairments and significantly improve system transmission performance.
[0072] This application also provides a first end-to-end inter-symbol interference compensation method, including: Step 10: Input the target data sequence into the transmitter processing module in the electronic device currently acting as the transmitter, so that the transmitter processing module outputs the target transmission signal corresponding to the target data sequence; It is understood that the target data sequence can be a binary bit sequence, and the target transmission signal refers to the light wave carrying information and the corresponding complex symbol sequence.
[0073] Step 20: The target transmission signal is transmitted to the receiving end, so that the receiving end receives the complex symbol sequence corresponding to the target transmission signal and inputs the complex symbol sequence into the receiving end processing module, thereby enabling the receiving end to obtain the interference compensation result data corresponding to the complex symbol sequence output by the receiving end processing module; wherein, the transmitting end processing module and the receiving end processing module constitute an inter-symbol interference compensation model for inter-symbol interference compensation of the data sequence, and the inter-symbol interference compensation model is pre-trained based on the training method of the end-to-end inter-symbol interference compensation model.
[0074] It is understood that the data sequence can be a binary bit sequence or a complex symbol sequence, and the interference compensation result data refers to the complex symbol sequence after inter-symbol interference compensation by the receiving end processing module.
[0075] This application also provides a second end-to-end inter-symbol interference compensation method, including: The electronic device currently acting as the receiving end receives the complex symbol sequence corresponding to the transmitted target signal and inputs the complex symbol sequence into the receiving end processing module in the electronic device, so that the receiving end processing module outputs the interference compensation result data corresponding to the complex symbol sequence; wherein, the target transmission signal is pre-input by the transmitting end into the transmitting end processing module, causing the transmitting end processing module to output the signal; wherein, the transmitting end processing module and the receiving end processing module constitute an inter-symbol interference compensation model for performing inter-symbol interference compensation on the received data sequence, and the inter-symbol interference compensation model is pre-trained based on the training method of the end-to-end inter-symbol interference compensation model.
[0076] It is understood that the target data sequence can be a binary bit sequence, the target transmission signal refers to the light wave carrying information and the corresponding complex symbol sequence, the data sequence can be a binary bit sequence or a complex symbol sequence, and the interference compensation result data refers to the complex symbol sequence after inter-symbol interference compensation by the receiving end processing module.
[0077] This application also provides an electronic device, which may include a processor, a memory, a receiver, and a transmitter. The processor is used to execute at least one of the training method for the end-to-end inter-symbol interference compensation model, the first end-to-end inter-symbol interference compensation method, and the second end-to-end inter-symbol interference compensation method mentioned in the above embodiments. The processor and the memory can be connected via a bus or other means, taking a bus connection as an example. The receiver can be connected to the processor and the memory via wired or wireless means.
[0078] The processor can be a central processing unit (CPU). The processor can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or combinations of the above types of chips.
[0079] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer-executable programs, and modules, such as at least one program instruction / module corresponding to the end-to-end inter-symbol interference compensation model training method, the first end-to-end inter-symbol interference compensation method, and the second end-to-end inter-symbol interference compensation method described in the embodiments of this application. The processor executes various functional applications and data processing by running the non-transitory software programs, instructions, and modules stored in the memory, thereby implementing at least one of the end-to-end inter-symbol interference compensation model training method, the first end-to-end inter-symbol interference compensation method, and the second end-to-end inter-symbol interference compensation method described in the above method embodiments.
[0080] The memory may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created by the processor, etc. Furthermore, the memory may include high-speed random access memory and non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory may optionally include memory remotely located relative to the processor, which can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.
[0081] The one or more modules are stored in the memory, and when executed by the processor, at least one of the training method of the end-to-end inter-symbol interference compensation model, the first end-to-end inter-symbol interference compensation method, and the second end-to-end inter-symbol interference compensation method described in the embodiment is executed.
[0082] In some embodiments of this application, the user equipment may include a processor, a memory, and a transceiver unit. The transceiver unit may include a receiver and a transmitter. The processor, memory, receiver, and transmitter may be connected via a bus system. The memory is used to store computer instructions, and the processor is used to execute the computer instructions stored in the memory to control the transceiver unit to send and receive signals.
[0083] As one implementation method, the functions of the receiver and transmitter in this application can be implemented by transceiver circuits or dedicated transceiver chips, and the processor can be implemented by dedicated processing chips, processing circuits or general-purpose chips.
[0084] As another implementation approach, the server provided in this application embodiment can be implemented using a general-purpose computer. That is, the program code implementing the processor, receiver, and transmitter functions is stored in memory, and the general-purpose processor implements the processor, receiver, and transmitter functions by executing the code in memory.
[0085] This application also provides a computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements at least one of the steps of the aforementioned training method for an end-to-end inter-symbol interference compensation model, the first end-to-end inter-symbol interference compensation method, and the second end-to-end inter-symbol interference compensation method. The computer-readable storage medium can be a tangible storage medium, such as random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, floppy disks, hard disks, removable storage disks, CD-ROMs, or any other form of storage medium known in the art.
[0086] This application also provides a computer program product, including a computer program that, when executed by a processor, implements at least one of the steps of the aforementioned training method for an end-to-end inter-symbol interference compensation model, the first end-to-end inter-symbol interference compensation method, and the second end-to-end inter-symbol interference compensation method.
[0087] Those skilled in the art will understand that the exemplary components, systems, and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software, or a combination of both. Whether implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application. When implemented in hardware, it can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. The programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave.
[0088] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.
[0089] In this application, features described and / or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, and / or combined with or in place of features of other embodiments.
[0090] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to the embodiments of this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.
Claims
1. A training method for an end-to-end inter-symbol interference compensation model, characterized in that, include: In the current iteration, the training data sequence is input into the current end-to-end system model used to simulate the super Nyquist coherent optical transmission system. This enables the transmitter processing module in the end-to-end system model to generate the simulated transmission signal corresponding to the training data sequence, and the channel module in the end-to-end system model to transmit the simulated transmission signal in a preset channel environment to obtain the received complex symbol sequence. The receiver processing module in the end-to-end system model then performs inter-symbol interference compensation on the received complex symbol sequence and outputs the interference compensation result data corresponding to the training data sequence. The target loss for the current iteration is determined based on the interference compensation result data, and the model parameters of the transmitter processing module and the receiver processing module are optimized based on the loss to update the end-to-end system model. If the end-to-end system model has converged or the current iteration round is the preset last iteration round, then the transmitter processing module and the receiver processing module in the updated end-to-end system model are determined as an inter-symbol interference compensation model for compensating for inter-symbol interference in the received data sequence.
2. The training method for the end-to-end inter-symbol interference compensation model according to claim 1, characterized in that, The transmitter processing module includes: The mapping module is used to perform QAM mapping on the pseudo-random code used as the training data sequence to obtain the mapping result data; The THP training module is used to pre-encode the mapping result data to obtain a pre-encoded complex symbol sequence; The Super Nyquist shaping module is used to shape the pre-encoded complex symbol sequence to obtain the simulated transmission signal.
3. The training method for the end-to-end inter-symbol interference compensation model according to claim 2, characterized in that, The THP training module includes: The first one-dimensional convolutional layer is used to filter the mapping result data to obtain the filtered first complex symbol sequence. The convolution kernel coefficients of the first one-dimensional convolutional layer are the tap coefficients to be trained in the THP training module. The first modulo operation unit is used to perform modulo operation on the filtered first complex number symbol sequence to generate a pre-encoded complex number symbol sequence.
4. The training method for the end-to-end inter-symbol interference compensation model according to claim 2, characterized in that, The super Nyquist shaping module includes: An upsampling unit is used to perform upsampling processing on the pre-encoded complex symbol sequence to obtain upsampling result data; The super Nyquist compression unit is used to perform super Nyquist compression on the upsampled result data to obtain compressed result data; A root-raised cosine filter is used to pulse-shape the compressed data to obtain a simulated transmission signal.
5. The training method for the end-to-end inter-symbol interference compensation model according to claim 2, characterized in that, The receiving end processing module includes an FFE training module, which includes: The second one-dimensional convolutional layer is used to filter the received complex symbol sequence to obtain the filtered second complex symbol sequence. The convolution kernel coefficients of the second one-dimensional convolutional layer are the tap coefficients to be trained in the FFE training module. The second modulo operation unit is used to perform modulo operation on the filtered second complex number symbol sequence to generate a repaired complex number symbol sequence.
6. The training method for the end-to-end inter-symbol interference compensation model according to claim 5, characterized in that, The step of determining the target loss for the current iteration based on the interference compensation result data, and optimizing the model parameters of the transmitting end processing module and the receiving end processing module based on the loss to update the end-to-end system model, includes: The mean square error between the interference compensation result data and the mapping result data is calculated. The mean square error is used as the loss function value of the end-to-end system model corresponding to the current iteration round. Based on the gradient optimization algorithm, the gradient of the model parameters is calculated according to the loss function value, and the model parameters are updated using the gradient. The model parameters include the tap coefficients of the THP training module and the tap coefficients of the FFE training module.
7. An end-to-end inter-symbol interference compensation method, characterized in that, include: The target data sequence is input into the transmitter processing module so that the transmitter processing module outputs the target transmission signal corresponding to the target data sequence. The target transmission signal is transmitted to the receiving end, so that the receiving end receives the complex symbol sequence corresponding to the target transmission signal and inputs the complex symbol sequence into the receiving end processing module, thereby enabling the receiving end to obtain the interference compensation result data corresponding to the complex symbol sequence output by the receiving end processing module. The transmitting end processing module and the receiving end processing module constitute an inter-symbol interference compensation model for compensating for inter-symbol interference in the received data sequence. The inter-symbol interference compensation model is pre-trained based on the training method of the end-to-end inter-symbol interference compensation model according to any one of claims 1 to 6.
8. An end-to-end inter-symbol interference compensation method, characterized in that, include: The receiver receives a complex symbol sequence corresponding to the target transmission signal and inputs the complex symbol sequence into the receiver processing module so that the receiver processing module outputs interference compensation result data corresponding to the complex symbol sequence; wherein, the target transmission signal is output by the transmitter processing module in advance by the transmitter inputting the target data sequence into the transmitter processing module. The transmitting end processing module and the receiving end processing module constitute an inter-symbol interference compensation model for compensating for inter-symbol interference in the received data sequence. The inter-symbol interference compensation model is pre-trained based on the training method of the end-to-end inter-symbol interference compensation model according to any one of claims 1 to 6.
9. An electronic device, characterized in that, It includes a processor and a memory; when the processor executes the running program stored in the memory, it implements at least one of the training method of the end-to-end inter-symbol interference compensation model as described in any one of claims 1 to 6, the end-to-end inter-symbol interference compensation method as described in claim 7, and the end-to-end inter-symbol interference compensation method as described in claim 8.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements at least one of the training methods for the end-to-end inter-symbol interference compensation model as described in any one of claims 1 to 6, the end-to-end inter-symbol interference compensation method as described in claim 7, and the end-to-end inter-symbol interference compensation method as described in claim 8.