Method and system for estimating transmission quality of ultra-wide-spectrum multi-band communication system

Abstract

According to the method and the system for estimating the transmission quality of the ultra-wide-spectrum multi-band communication system, a CF-ISRS-GN model is used for estimating the transmission quality of the system under different transmission conditions, and compared with a method for estimating the transmission quality based on an SSFM algorithm, an estimation error generated by the CF-ISRS-GN model is obtained; establishing a data set required by a correction model; and correcting the CF-ISRS-GN model by using a neural network, training the neural network by taking the optical fiber link parameter, the channel parameter and the SNR obtained by the CF-ISRS-GN model as characteristic values and setting the estimation error as a label value, and finally obtaining a neural network model with the CF-ISRS-GN model correction capability. Compared with a traditional transmission quality estimation scheme, the method has the advantages that the model is reliable, explainable and generalizable, and the estimation precision of the transmission quality can be further improved by correcting the model through the correction model.

Description

Technical Field

[0001] This invention proposes a method and system for estimating the transmission quality of an ultra-wideband multi-band communication system, which relates to the research field of measuring system transmission quality in optical fiber communication technology. Background Technology

[0002] The rapid growth of 5G services has placed higher demands on optical network capacity, leading to the deployment of unused low-loss frequency bands in optical fibers to expand network capacity. Currently, there is a trend towards deploying C+L band optical communication systems to accommodate the surge in data traffic. However, in multi-band systems, compared to traditional C-band systems, power transfer caused by inter-channel stimulated Raman scattering (ISRS) becomes more significant. This reliance on channel transmit power distribution complicates the estimation of nonlinear interference, further hindering accurate estimation of system transmission quality. Modeling ultra-wideband multi-band systems allows for optimization of communication links to maximize system capacity. Current methods for transmission quality estimation based on system modeling include schemes based on the Split-step Fourier Method (SSFM), schemes based on Gaussian noise (GN) models, and machine learning-based schemes.

[0003] Gaussian noise models are widely used for rapid evaluation of nonlinear interference (NLI) noise, and together with amplified spontaneous emission (ASE) noise, they define the system signal-to-noise ratio (SNR) to measure the transmission quality of the system. With the expansion of fiber optic frequency bands, to estimate the impact of ISRS on Kerr nonlinearity, the Inter-channel Stimulated Raman Scattering-Gaussian Noise (ISRS-GN) model was proposed. Its principle is based on the GN model, estimating the Kerr nonlinearity by evaluating the power distribution of arbitrary signals along the link between different channels. To further accelerate the estimation, a closed-form ISRS-GN (CF-ISRS-GN) model was proposed. In its derivation, a triangular approximation was applied to the Raman gain spectrum. Assumptions were made that ISRS has a minimal impact on channel power distribution, that channel power attenuation is solely a function of total transmit power and independent of its spectral distribution, and that the frequency interval between two channels is much greater than half the channel bandwidth. This resulted in the power spectral density of the closed-form nonlinear interference, enabling transmission quality estimation for transmission systems with bandwidths approaching 10 THz. In summary, achieving fast transmission quality estimation using the GN model requires sacrificing estimation accuracy, and the errors caused by the model's theoretical assumptions are difficult to analyze completely and accurately. With the rise of neural networks, data-driven machine learning methods have become an alternative for fiber optic channel modeling and transmission quality estimation. However, these methods rely on the data itself, lack interpretability, and cannot be universally adapted to various communication systems with different transmission conditions. Therefore, it is necessary to further improve the accuracy and applicability of transmission quality estimation. Commonly used methods include combining analytical models and machine learning. A scheme is proposed to modify the ISRS-GN model to achieve transmission quality estimation. This scheme utilizes the superior nonlinear fitting properties of neural networks to modify the ISRS-GN model and thus achieve accurate transmission quality estimation.

[0004] This invention proposes a method and system for estimating transmission quality in ultra-wideband multi-band communication systems based on the aforementioned requirements. It utilizes the CF-ISRS-GN model to estimate the transmission quality of the system under different transmission conditions, and compares this estimation with a method based on the SSFM algorithm to obtain the estimation error generated by the former, thereby establishing the dataset required for the corrected model. A neural network is then used to correct the CF-ISRS-GN model. The neural network is trained using fiber link parameters, channel parameters, and the SNR obtained from the CF-ISRS-GN model as feature values, and the estimation error as the label value, ultimately resulting in a neural network model capable of correcting the CF-ISRS-GN model. Compared to traditional transmission quality estimation schemes, this invention utilizes the CF-ISRS-GN model, enabling the model to possess reliability, interpretability, and generalization capabilities. Furthermore, correcting the model with a modified version further improves the accuracy of transmission quality estimation. Summary of the Invention

[0005] This invention proposes a method and system for estimating transmission quality in an ultra-wideband multi-band communication system, the system structure of which is as follows: Figure 1 As shown, it consists of a transmitter, an optical fiber link, a receiver, and a digital signal processing system, wherein:

[0006] Transmitter: Used to generate optical signals of different bands and channels. Multiple optical signals from multiple channels are multiplexed into one channel by a multiplexer to obtain a wavelength division multiplexing (WDM) signal, which is then transmitted into the optical fiber.

[0007] Optical fiber links are used to transmit optical signals. Optical signals are susceptible to damage within the link, primarily consisting of three parts: attenuation loss, dispersion damage, and nonlinear effects. Stimulated Raman scattering between channels occurs within the optical fiber channel and is a type of nonlinear effect damage. This effect causes power transfer between different channels, thus affecting the signal transmission quality in the optical fiber.

[0008] Receiver: The WDM optical signal is demultiplexed into various bands by a demultiplexer, and then the optical signal transmitted by the channel under test is obtained by a bandpass filter. The optical signal is converted into an electrical signal and sampled at a certain sampling frequency. The sampled value is converted into the corresponding digital code to obtain a digital signal, which is then transmitted to the digital signal processing system for processing.

[0009] The digital signal processing system (DSP) is used to process signals and includes modules for dispersion compensation, IQ imbalance compensation, clock recovery, adaptive equalization, carrier phase recovery, and digital demodulation. In the DSP, the dispersion compensation module compensates for signal impairments, and then the signal reaches the IQ imbalance compensation module to handle the orthogonality issue. The clock signal is extracted from the compensated signal in the clock recovery module and then enters the carrier phase module to eliminate phase noise. The processed signal enters the digital demodulation module for digital demodulation. After demodulation, the signal is output as a decision signal and compared with the transmitted signal to calculate the EVM value. The relationship between EVM and SNR is used to obtain the SNR of the current channel, which serves as an indicator of the channel's transmission quality. The functions of each module are as follows:

[0010] Dispersion compensation module: compensates for the dispersion damage that the signal suffers during optical fiber transmission.

[0011] IQ Imbalance Compensation Module: The signal is processed using the Generalized Subspace Orthogonalization and Projection (GSOP) algorithm.

[0012] Clock recovery module: Extracts the clock signal from the signal based on the reference clock.

[0013] Adaptive equalization module: Estimates the frequency difference between the received digital signal and the local oscillator light source, and compensates for the frequency deviation.

[0014] Carrier phase recovery module: Eliminates significant phase noise imposed on the signal by frequency offset and phase shift.

[0015] Digital demodulation module: Demaps the compensated digital signal into a bit sequence.

[0016] Based on the above simulated transmission system, this invention proposes a method for estimating the transmission quality of an ultra-wideband multi-band communication system, the block diagram of which is shown below. Figure 2 As shown, it includes the following steps:

[0017] Step 1: Obtain the signal-to-noise ratio (SNR) 1 based on the distributed Fourier transform SSFM algorithm through the simulation system, and use it as the standard for measuring transmission quality estimation.

[0018] Step 2: Collect system and channel parameters from the simulation system, including fiber optic link parameters (attenuation coefficient, dispersion coefficient, second-order dispersion coefficient, third-order dispersion coefficient, nonlinear coefficient, length per fiber span, total fiber length, Raman gain slope, a total of 9) and channel parameters (channel number, transmit power, channel bandwidth, a total of 3).

[0019] Step 3: Input the collected system and channel parameter information into the closed-form inter-channel stimulated Raman scattering-Gaussian noise CF-ISRS-GN model to obtain SNR2 based on the CF-ISRS-GN model, and calculate the error value with SNR1.

[0020] Step 4: To obtain the neural network correction model, it is necessary to simulate different transmission conditions to build the dataset required for the neural network model. By changing the fiber optic link parameters in the simulation system and estimating the transmission quality of different channels, repeat steps 1-3 to build the dataset.

[0021] Step 5: Use SNR2 and the collected system and channel parameter information as channel feature vectors, and use the error generated by SNR1 and SNR2 as label values ​​to train the neural network correction model. Finally, select the model with the smallest mean square error in the cross-validation process as the neural network correction model.

[0022] Step 6: After the neural network model is established, only the SNR2 obtained from the CF-ISRS-GN model and the system and channel parameters are required. The error ΔSNR generated by the CF-ISRS-GN model can be predicted by the neural network correction model, which is then used to correct SNR2 and finally obtain the corrected SNR as a measure of the transmission quality of the channel under test.

[0023] The following provides a more detailed explanation of each step:

[0024] In step 1, after obtaining the EVM value of the channel under test through the digital signal processing module of the simulation system, the value is then processed... SNR1 is calculated, where N is the total number of bits in the signal, and R(j) and T(j) are the bit sequences at the receiving end and the transmitting end, respectively.

[0025] In step 2, the fiber optic link parameters (attenuation coefficient α, dispersion coefficient D, second-order dispersion coefficient β2, third-order dispersion coefficient β3, nonlinear coefficient γ, and fiber length per span L) are specified. s Number of segments N s Total fiber length L (i.e., L = N) s L s Raman gain slope C r (9 in total) Channel parameters (channel number Num, transmit power P) i (or P) k ), channel bandwidth B i (or B) k (3 in total), which are used to calculate the nonlinear interference coefficient of the Ns segment transmitted in the i-th channel via CF-ISRS-GN. And as part of the input feature values ​​for the corrected neural network.

[0026] In step 3, the 12 parameters from step 2 are used... To calculate the SNR² of the i-th channel, the CF-ISRS-GN model is used to calculate the nonlinear interference coefficient of the i-th channel. SNR2, along with 12 parameters, is used as the input feature value of the corrected neural network. The error of SNR is calculated by ΔSNR = SNR2 - SNR1, and this error will be used as the output label value of the corrected neural network.

[0027] In step 4, the length L of each fiber span in the fiber optic link is modified in the simulation system. s Number of segments N s And modify the transmit power P of each channel. i (or P) k This method simulates different transmission conditions and obtains the SNR1 and SNR2 of different channels as test channels, thus estimating the transmission quality of the channel and building a dataset. The neural network model block diagram is shown below. Figure 3 As shown.

[0028] In step 5, the algorithm for training the neural network correction model is shown in Table 1. First, the neural network model is initialized according to the selected hyperparameters, and a bias and weight parameter is randomly selected for each neuron in the layer. Next, the feature values ​​of the established dataset and their corresponding label values ​​are input into the model, and the estimated values ​​under these input feature values ​​are calculated. Then, the gradient descent algorithm is used to calculate the gradient of the loss function to further update the weight parameters. Finally, the model with the weight parameters that minimize the loss function is selected as the neural network correction model. ΔSNR is used as the label value, and it is compared with the ΔSNR estimated by the neural network model. pred The loss function is calculated, and the weight parameters in the neural network are updated according to the backpropagation algorithm to train the model; when the mean squared error is minimized, i.e.

[0029] in,

[0030] f loss =MSE(ΔSNR,ΔSNR) pred ), where ΔSNR is the SNR error calculated from SNR1 and SNR2, ΔSNR pred The SNR estimated by the neural network correction model.

[0031] In step 6, the attenuation coefficient α, dispersion coefficient D, second-order dispersion coefficient β2, third-order dispersion coefficient β3, nonlinear coefficient γ, and fiber length L per span are provided under the current transmission conditions. s Number of segments Ns Total fiber length L, Raman gain slope C r Channel number Num, transmit power P i Channel bandwidth B i And the SNR2 obtained from the CF-ISRS-GN model (a total of 13 parameters, using...) (to represent the input feature vector) is fed into the correction neural network, and then processed by the correction neural network. (That is, the trained neural network is theoretically a mapping function f) to obtain ΔSNR. pred Ultimately, it is achieved through SNR = SNR² - ΔSNR pred The corrected SNR is obtained.

[0032] Table 1 Algorithm for Training Neural Network and Correcting the Model

[0033]

[0034] The advantages and beneficial effects of this invention are as follows:

[0035] To address the limitation of accuracy in transmission quality estimation using numerical analysis models in ultra-wideband multi-band optical communication systems, this invention proposes a method and system for transmission quality estimation in such systems. Utilizing the superior nonlinear fitting and learning capabilities of neural networks, the error between SNR estimations based on the CF-ISRS-GN model and the SSFM model is predicted, thereby correcting the CF-ISRS-GN model. Compared to traditional methods, this invention enables rapid and accurate transmission quality estimation of the channel under test in a multi-band optical communication system, providing a reliable and accurate transmission quality index for optical communication system management and optimization. Attached Figure Description

[0036] Figure 1 This is a block diagram of an ultrawideband multi-band simulated optical communication system.

[0037] Figure 2 This is a block diagram of the transmission quality estimation method.

[0038] Figure 3 This is a block diagram of a neural network model.

[0039] Figure 4a , Figure 4b , Figure 4c , Figure 4d , Figure 4e This is an error correction for the CF-ISRS-GN model based on a neural network model for systems with different transmit power at the same transmission distance.

[0040] Figure 5a , Figure 5b , Figure 5c , Figure 5d , Figure 5e , Figure 5f This is an error correction for the CF-ISRS-GN model based on a neural network model for systems with different transmission distances under the same transmit power. Detailed Implementation

[0041] The method and system for estimating transmission quality in an ultra-wideband multi-band communication system proposed in this invention can be applied to ultra-wideband multi-band optical communication systems. The workflow of a specific embodiment is as follows:

[0042] Construction of an ultrawideband multi-band optical communication system:

[0043] (1) Transmitter: Each channel is equipped with a laser to generate optical signals with different carrier frequencies. The carrier frequency range of the S-band is 197.325THz-203.025THz, the frequency range of the C-band is 190.975THz-196.675THz, and the frequency range of the L-band is 184.625THz-190.325THz. The channel spacing is 300GHz, and there is also a guard interval between different bands, so a total of 57 channels are generated. The 256GBaud DP-16QAM modulated signal is generated by an arbitrary waveform generator (AWG) and drives the IQ modulator to generate a high-speed single-carrier optical signal for each channel.

[0044] (2) Fiber optic link: Single-mode fiber is used as the transmission fiber with a center wavelength of 1550nm, loss of 0.2dB / km, dispersion of 16.75ps / nm / km, nonlinear coefficient of 1.3 / w / km, and Raman gain slope of 0.028 / w / km / THz. The length and number of each span of the fiber optic link can be modified to simulate different transmission conditions. A power control mode amplifier is selected to compensate for the attenuated and power-transferred optical signal, with a noise figure of 5dB.

[0045] (3) Receiver: The WDM optical signal is demultiplexed into various bands by a demultiplexer, and then the optical signal carrying the modulation signal of the channel under test is obtained by a bandpass filter. In the coherent receiver, the optical signal is converted into an electrical signal by a photodetector. The electrical signal is converted into a digital signal by analog-to-digital conversion, and then processed in the digital signal processing module. Finally, the corresponding digital demodulation is performed according to the modulation method of the signal to obtain the bit sequence. Then, the EVM of the signal is calculated and further converted into SNR as a measure of the transmission quality of the signal.

[0046] Transmission quality estimation using the CF-ISRS-GN model:

[0047] (1) Extract parameters of multi-band system and fiber optic link: record channel bandwidth, number of channels, channel carrier frequency, channel transmit power; record attenuation coefficient, dispersion coefficient, nonlinear coefficient, Raman gain slope, number of spans, span length of fiber optic link; record the noise figure of amplifier.

[0048] (2) Calculate spontaneous radiated noise power: through To calculate, where P ASE (f i Let N be the spontaneously radiated noise power of the i-th channel (i.e., the channel under test), and N be the radiated noise power of the i-th channel. s Let f be the total number of segments in the transmission fiber optic link, h be Planck's constant, and f be the total number of segments in the transmission fiber optic link. i Let be the carrier frequency of the i-th channel, NF be the noise figure of the amplifier, and G be the carrier frequency of the i-th channel. n B is the gain value of the nth segment amplifier. i Let be the modulation bandwidth of the i-th channel.

[0049] (3) Calculate the nonlinear interference noise power: The definition of nonlinear interference noise power in the CF-ISRS-GN model is... in, This represents the nonlinear interference noise figure. Using the CF-ISRS-GN model, the nonlinear noise figure can be defined as... in It is the nonlinear disturbance coefficient term calculated by the CF-ISRS-GN model. It is a modulation format correction item.

[0050] For the nonlinear disturbance coefficient term calculated by the CF-ISRS-GN model, Among them, P i,j Let P be the power transmitted from the i-th channel to the j-th segment, and when j=1, P i,1 =P i ε is the accumulation parameter for coherent or incoherent SPM. When ε = 0, it is coherent accumulation; when ε = 0, it is incoherent accumulation. When, it is an incoherent accumulation, where L s Where β is the single span length of the optical fiber, β2 is the second-order dispersion coefficient, and B i For channel bandwidth, N ch Δf represents the total number of channels and the channel interval.

[0051] for The first term is the nonlinear interference coefficient generated by SPM in the j-th segment:

[0052]

[0053] Among them, f iLet γ be the carrier frequency of the i-th channel (i.e., the channel under test), γ be the fiber nonlinearity coefficient, and B be the carrier frequency of the i-th channel (i.e., the channel under test). i For channel bandwidth, β2 and β3 are the second-order and third-order dispersion coefficients of the optical fiber, respectively. and The fiber loss factor is... P tot C is the sum of the transmit power of all channels. r This represents the Raman gain slope coefficient.

[0054] The second term is the XPM nonlinear disturbance coefficient generated in the j-th span:

[0055]

[0056] Among them, f i Let N be the carrier frequency of the i-th channel (i.e., the channel under test). ch P represents the total number of channels. i and P k These represent the transmit power of the i-th and k-th channels, respectively, where γ is the fiber nonlinearity coefficient, and B... i and B k φ represents the channel bandwidth of the i-th channel and the k-th channel, respectively. i,k =2π 2 (f k -f i )[β2+πβ3(f i +f k )], β2 and β3 are the second-order and third-order dispersion coefficients of the optical fiber, respectively, f k Let k be the carrier frequency of the k-th channel. and The fiber loss factor is... P tot C is the sum of the transmit power of all channels. r This represents the Raman gain slope coefficient.

[0057] For the modulation format correction term, it is defined as:

[0058]

[0059] Among them, f i Let N be the carrier frequency of the i-th channel (i.e., the channel under test). ch P represents the total number of channels. i and P k These represent the transmit power of the i-th and k-th channels, respectively, where γ is the fiber nonlinearity coefficient, and B... i and B k The channel bandwidths of the i-th and k-th channels are respectively, φ = -4π. 2[β2+πβ3(f i +f k )]L s φ i,k =2π 2 (f k -f i )[β2+πβ3(f i +f k )], L s f is the fiber span length, β2 and β3 are the second and third order dispersion coefficients of the fiber, respectively, and f k Let k be the carrier frequency of the k-th channel. and The fiber loss factor is... P tot C is the sum of the transmit power of all channels. r The slope coefficient of the Raman gain. For cross-segment related parameters: when N s When = 1, When N s When ≠1,

[0060] (4) Calculate the SNR2 of the channel under test: In a multi-band transmission system, after coherent detection and dispersion compensation, the channel-related SNR2 is defined as...

[0061] Neural network dataset creation: In order to simulate the transmission system under different transmission conditions and estimate its transmission quality, a dataset was generated by modifying the transmit power and fiber optic link length. The specific data information is shown in Tables 1 and 2.

[0062] Table 1. Dataset-related parameters

[0063]

[0064] Table 2 Dataset Related Parameters

[0065]

[0066] Neural network correction model establishment:

[0067] (1) Selection of feature values ​​and label values: Fiber link parameters (attenuation coefficient, dispersion coefficient, second-order dispersion coefficient, third-order dispersion coefficient, nonlinear coefficient, length per fiber span, total fiber length, Raman gain slope, a total of 9), channel parameters (channel number, transmit power, channel bandwidth, a total of 3), and SNR2 obtained by using the CF-ISRS-GN model are selected as 13 input feature values ​​of the neural network. The error ΔSNR between SNR1 obtained based on SSFM and SNR2 obtained based on the CF-ISRS-GN model is used as 1 label value of the output layer.

[0068] (2) Determination of Neural Network Model Parameters: The hyperparameters of the neural network were determined using the grid method. The number of neurons in the input layer was the same as the input feature values, and the number of neurons in the output layer was the same as the label values. The number of neurons in the hidden layer was set to N = {16, 32, 64, 128, 256, 512}, the number of hidden layers was Layer = {1, 2, 3, 4}, the learning rate was Lr = {0.001:0.001:0.01}, the batch learning parameter was b = {16, 32, 64, 128}, and the number of iterations was e = {100, 200, 300, 400}. The activation function was Tanh to ensure that the features maintained the same sign during transformation, corresponding to the positive and negative signs of the error. At the same time, the z-score method was used for feature preprocessing to ensure that the output variation range was corresponding. All neural networks were trained on the established dataset, and the neural network with the smallest average error during cross-validation was selected as the error correction model. The final network structure is shown in Table 3.

[0069] Table 3 Neural Network Model Parameters

[0070]

[0071] Transmission quality estimation method: After establishing a modified model based on neural networks, channel information, fiber optic link information, and SNR estimated based on the CF-ISRS-GN model can be used to achieve reliable and accurate transmission quality estimation for each channel of the transmission system.

[0072] This invention also provides a method for estimating transmission quality in a multi-band transmission system. To verify the accuracy of the proposed model under specific transmission conditions, and considering that ISRS strength is related to channel transmit power and transmission distance, system conditions of the same transmission distance but different transmit power and the same transmit power but different transmission distances are selected for analysis. Figures 4a-4e , Figures 5a-5f As shown.

[0073] like Figures 4a-4eThis paper demonstrates the performance of a 256GBd DP-16QAM signal transmitted through four 100km standard single-mode fiber optic links. The channel transmit power was modified from 5dBm to 9dBm, and the ISRS increased with increasing transmit power. The SNR before and after the correction was compared with the SNR derived from SSFM to verify the correction of the neural network model for CF-ISRS-GN. The results show that under different transmit power conditions, the corrected SNR (square icon) is closer to the SNR derived from SSFM (cross-shaped icon) than the uncorrected SNR (triangle icon). This indicates that the model proposed in this invention has better transmission quality estimation performance. Figures 5a-5f As shown, with a channel transmit power of 7dBm and transmission distances of 100km, 200km, 300km, 400km, 500km, and 600km, nonlinear interference accumulates as the transmission distance increases. By comparing the SNR before and after correction with the SNR derived from SSFM, the model is verified to have good transmission quality estimation performance for transmission systems at different distances.

Claims

1. A transmission quality estimation system for an ultra-wideband multi-band communication system, comprising a transmitter, an optical fiber link, a receiver, and a digital signal processing system, characterized in that: Transmitter: Used to generate optical signals of different bands and channels. Multiple optical signals from multiple channels are multiplexed into one channel by a multiplexer to obtain a wavelength division multiplexing (WDM) signal, which is then transmitted into the optical fiber. Fiber optic link: used to transmit optical signals; Receiver: The WDM optical signal is demultiplexed into various bands by a demultiplexer, and then the optical signal transmitted by the channel under test is obtained by a bandpass filter. The optical signal is converted into an electrical signal and sampled by a sampling frequency. The sampled value is converted into the corresponding digital code to obtain a digital signal, which is then transmitted to the digital signal processing system for processing. Digital signal processing system: used to process signals, including dispersion compensation module, IQ imbalance compensation module, clock recovery module, adaptive equalization module, carrier phase recovery module, and digital demodulation module; in the digital signal processing system, the dispersion compensation module is used to compensate for signal impairments, and then the signal reaches the IQ imbalance compensation module to handle the orthogonality problem of the signal; The clock signal is extracted from the clock recovery module after compensation, and then enters the carrier phase module to eliminate phase noise. The processed signal enters the digital demodulation module for digital demodulation. After demodulation, the signal is output to the decision output module to recover the signal. It is compared with the signal at the transmitting end to calculate the EVM value. Through the relationship between EVM and SNR, the SNR of the current channel is obtained as an indicator to measure the transmission quality of the channel.

2. The transmission quality estimation system for an ultra-wideband multi-band communication system according to claim 1, characterized in that: The dispersion compensation module is used to compensate for the dispersion damage suffered by the signal during optical fiber transmission; the IQ imbalance compensation module uses the orthogonal imbalance algorithm GSOP to process the signal. The clock recovery module extracts the clock signal from the signal based on the reference clock; the adaptive equalization module estimates the frequency difference between the received digital signal and the local oscillator and compensates for the frequency deviation; the carrier phase recovery module eliminates the significant phase noise imposed on the signal by frequency offset and phase shift; and the digital demodulation module demodulates the compensated digital signal into a bit sequence.

3. A method for estimating transmission quality in an ultra-wideband multi-band communication system, characterized in that, Includes the following steps: Step 1: Obtain the signal-to-noise ratio (SNR) 1 based on the distributed Fourier transform SSFM algorithm through the simulation system, and use it as the standard for measuring transmission quality estimation; Step 2: Collect system and channel parameters from the simulation system, including fiber optic link parameters and channel parameters; Step 3: Input the collected system and channel parameter information into the closed-form inter-channel stimulated Raman scattering-Gaussian noise CF-ISRS-GN model to obtain SNR2 based on the CF-ISRS-GN model, and calculate the error value with SNR1 at the same time. Step 4: By changing the fiber optic link parameters in the simulation system and estimating the transmission quality of different channels, repeat steps 1-3 to establish a dataset. Step 5: Use SNR2 and the collected system and channel parameter information as channel feature vectors, use the error generated by SNR1 and SNR2 as label values, train the neural network correction model, and finally select the model with the smallest mean square error in the cross-validation process as the neural network correction model. Step 6: After the neural network model is established, it is only necessary to provide the SNR2 obtained from the CF-ISRS-GN model and the system and channel parameters. That is, the error ΔSNR generated by the CF-ISRS-GN model is predicted by the neural network correction model, which is then used to correct SNR2, and finally the corrected SNR is obtained as a measure of the transmission quality of the channel under test.

4. The method for estimating transmission quality in an ultra-wideband multi-band communication system according to claim 3, characterized in that: In step 1, after obtaining the EVM value of the channel under test through the digital signal processing module of the simulation system, the value is then processed... SNR1 is calculated, where N is the total number of bits in the signal, and R(j) and T(j) are the bit sequences at the receiving end and the transmitting end, respectively.

5. The method for estimating transmission quality in an ultra-wideband multi-band communication system according to claim 3, characterized in that: In step 2, the fiber optic link parameters include: attenuation coefficient α, dispersion coefficient D, second-order dispersion coefficient β2, third-order dispersion coefficient β3, nonlinear coefficient γ, and fiber length per span L. s Number of segments N s The total fiber length L, i.e., L = N s L s Raman gain slope C r Channel parameters include: channel number Num, transmit power P i Or P k Channel bandwidth B i Or B k ; Used to calculate the nonlinear interference coefficient of the Ns segment transmitted in the i-th channel via CF-ISRS-GN And as part of the input feature values ​​for the corrected neural network.

6. A method for estimating transmission quality in an ultra-wideband multi-band communication system according to claim 3 or 5, characterized in that: In step 3, using To calculate the SNR² of the i-th channel, the nonlinear interference coefficient of the i-th channel is calculated using the CF-ISRS-GN model. SNR2, along with fiber optic link parameters and channel parameters, are used as input feature values ​​for the corrected neural network. The error of SNR is calculated using ΔSNR = SNR2 - SNR1, and this error is used as the output label value of the corrected neural network.

7. The method for estimating transmission quality in an ultra-wideband multi-band communication system according to claim 3, characterized in that: In step 4, the length L of each fiber span in the fiber optic link is modified in the simulation system. s Number of segments N s And modify the transmit power P of each channel. i Or P k To simulate different transmission conditions, different channels are selected as the channels to be tested to obtain the channel's SNR1 and SNR2, that is, to estimate the transmission quality of the channel, thereby establishing a dataset.

8. The method for estimating transmission quality in an ultra-wideband multi-band communication system according to claim 3, characterized in that: In step 5, the neural network model is first initialized according to the selected parameters, and a bias and weight parameter is randomly selected for each neuron in the layer. Then, the feature values ​​of the established dataset and their corresponding label values ​​are input into the model, and the estimated values ​​corresponding to the input feature values ​​are calculated. Next, the gradient descent algorithm is used to calculate the gradient of the loss function to further update the weight parameters. Finally, the model with the weight parameters that minimize the loss function is selected as the neural network correction model.

9. The method for estimating transmission quality in an ultra-wideband multi-band communication system according to claim 8, characterized in that: Using ΔSNR as the label value, and comparing it with the ΔSNR estimated by the neural network model... pred The loss function is calculated, and the weight parameters in the neural network are updated according to the backpropagation algorithm to train the model; when the mean squared error is minimized, i.e. Among them, f loss =MSE(ΔSNR,ΔSNR) pred ), where ΔSNR is the SNR error calculated from SNR1 and SNR2, ΔSNR pred The SNR estimated by the neural network correction model.

10. The method for estimating transmission quality in an ultra-wideband multi-band communication system according to claim 3, characterized in that: In step 6, the attenuation coefficient α, dispersion coefficient D, second-order dispersion coefficient β2, third-order dispersion coefficient β3, nonlinear coefficient γ, and fiber length L per span are provided under the current transmission conditions. s Number of segments N s Total fiber length L, Raman gain slope C r Channel number Num, transmit power P i Channel bandwidth B i The SNR2 obtained from the CF-ISRS-GN model is fed into the corrected neural network, and then processed by the corrected neural network f: To obtain ΔSNR pred Ultimately, it is achieved through SNR = SNR² - ΔSNR pred The corrected SNR is obtained.

Images